{
  "asOf": "2026-09-04",
  "rows": [
    {
      "name": "OpenAI",
      "slug": "openai",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party model lab; the default frontier API for GPT-5 family text, image, audio and embeddings.",
      "offering": "Frontier proprietary API (GPT-6 Astra, GPT-5.6 Sol/Terra/Luna, GPT-5.5, GPT-5.4/Mini/Nano, GPT-5.x Pro, GPT-5.3 Codex), plus Batch and Realtime APIs.",
      "pricingModel": "per-token",
      "representativePricing": "GPT-6 Astra $10/$50 ($1 cached input; 2x on prompts over 272K tokens, where output is $75; Fast mode 2x standard). GPT-5.6 Sol $4/$20 (promotional through at least Nov 21, 2026; was $5/$30), Terra $2/$12, Luna $0.20/$1.20; GPT-5.5 $5/$30, Nano $0.20/$1.25 per Mtok. Batch and Flex 50% off; cached input up to 90% off.",
      "standout": "GPT-6 Astra, released Sept 3, 2026, is the new flagship at $10/$50 per Mtok, 2.5x the GPT-5.6 Sol rate it replaces and level with Claude Fable 5.1. The sticker overstates the bill: Astra tops the independent Terminal-Bench 4.0 board at 58.18% and solves a task for $17.02 against Fable 5.1's $32.69, because it burns 1.53B tokens where Fable 5.1 burns 2.75B. On Artificial Analysis it is less dominant, tying its own predecessor at Intelligence Index 61. Note: winding down its fine-tuning platform (closed to new users as of May 2026).",
      "flagshipModel": "GPT-6 Astra",
      "flagshipOutputPerMtok": 50,
      "openAICompatible": true,
      "freeTier": "$5 in free credits for new accounts (expire in 3 months).",
      "homeUrl": "https://openai.com",
      "pricingUrl": "https://developers.openai.com/api/docs/pricing",
      "docsUrl": "https://platform.openai.com",
      "sourceUrl": "https://developers.openai.com/api/docs/pricing",
      "verifiedDate": "2026-09-04",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Anthropic",
      "slug": "anthropic",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party model lab focused on safety, coding and agentic work; the Claude family.",
      "offering": "Frontier proprietary API: Claude Fable 5.1 (from September 1, 2026) at the top, then Opus 5, Sonnet 5 and Haiku 4.5, with Mythos 5.1 restricted to Project Glasswing. All at 1M-token context.",
      "pricingModel": "per-token",
      "representativePricing": "Fable 5.1 $10/$50 with a $0.25 cache read, Opus 5 $5/$25, Sonnet 5 $2/$10 (the introductory rate made permanent), Haiku 4.5 $1/$5 per Mtok. Batch 50% off; cache reads cost 10% of input, or 2.5% on Fable 5.1.",
      "standout": "Output costs 5x input across the line; pricing has held steady across generations, with Fable 5.1 cutting the cache read rather than the sticker. Run-rate revenue passed $30B in early 2026.",
      "flagshipModel": "Claude Fable 5.1",
      "flagshipOutputPerMtok": 50,
      "openAICompatible": false,
      "freeTier": "",
      "homeUrl": "https://anthropic.com",
      "pricingUrl": "https://claude.com/pricing",
      "docsUrl": "https://platform.claude.com/docs",
      "sourceUrl": "https://platform.claude.com/docs/en/about-claude/pricing",
      "verifiedDate": "2026-09-02",
      "status": "representative",
      "note": "US-only inference adds a 1.1x multiplier."
    },
    {
      "name": "Google",
      "slug": "google",
      "category": "first-party-lab",
      "region": "US",
      "summary": "Hyperscaler and first-party lab; Gemini via the Gemini API, AI Studio and Vertex AI.",
      "offering": "Gemini 3.x API (3.1 Pro, 3.5 Flash, 3.1 Flash-Lite). AI Studio is free for prototyping; Vertex AI adds enterprise SLAs and compliance.",
      "pricingModel": "per-token",
      "representativePricing": "Gemini 3.1 Pro $2/$12 (to 200K ctx), 3.5 Flash $1.50/$9, Flash-Lite $0.25/$1.50 per Mtok. 90% context-caching discount.",
      "standout": "Largest production context window (2M tokens) and the cheapest Tier-1 budget model (Flash-Lite). Pro models are paid-only as of April 1, 2026.",
      "flagshipModel": "Gemini 3.1 Pro",
      "flagshipOutputPerMtok": 12,
      "openAICompatible": true,
      "freeTier": "AI Studio free for prototyping; Flash retains a free tier.",
      "homeUrl": "https://ai.google.dev",
      "pricingUrl": "https://ai.google.dev/gemini-api/docs/pricing",
      "docsUrl": "https://cloud.google.com/vertex-ai",
      "sourceUrl": "https://ai.google.dev/gemini-api/docs/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "xAI (Grok)",
      "slug": "xai-grok",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party model lab; Grok, the only frontier model with live grounding to X posts.",
      "offering": "Frontier proprietary API: Grok 4.5 (flagship, 500K ctx), Grok 4.20 (2M ctx long-context), Grok 4.1 Fast (cheap workhorse).",
      "pricingModel": "per-token",
      "representativePricing": "Grok 4.5 $2/$6, Grok 4.1 Fast $0.20/$0.50 per Mtok. Batch 50% off; cached input ~90% off.",
      "standout": "Only frontier model with live grounding to X posts; xAI markets Grok 4.5 (released July 8, 2026) as Opus-class at $2/$6, trained jointly with Cursor and priced well below Opus 4.8 and GPT-5.6 Sol. API is independent of X subscriptions.",
      "flagshipModel": "Grok 4.5",
      "flagshipOutputPerMtok": 6,
      "openAICompatible": true,
      "freeTier": "Free developer credits via data-sharing program (~$150-175/mo reported).",
      "homeUrl": "https://x.ai",
      "pricingUrl": "https://x.ai/api",
      "docsUrl": "https://docs.x.ai",
      "sourceUrl": "https://x.ai/api",
      "verifiedDate": "2026-07-10",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Meta",
      "slug": "meta",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party model lab (Meta Superintelligence Labs); the Muse Spark Model API is Meta's first paid, closed frontier model, separate from the open-weight Llama line.",
      "offering": "First-party Model API in public preview: Muse Spark 1.1, a multimodal agentic reasoning model with a 1M-token context (Meta-reported). OpenAI-compatible request format.",
      "pricingModel": "per-token",
      "representativePricing": "Muse Spark 1.1 $1.25/$4.25 per Mtok. No cached-input rate published yet.",
      "standout": "Meta's first paid model and its move from open-weight-only to a closed frontier API; leads tool-use and agentic tests on Meta's own numbers but trails the coding frontier. US-only and waitlisted at launch, and kept off third-party platforms.",
      "flagshipModel": "Muse Spark 1.1",
      "flagshipOutputPerMtok": 4.25,
      "openAICompatible": true,
      "freeTier": "$20 in free credits for new API accounts.",
      "homeUrl": "https://ai.meta.com",
      "pricingUrl": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/",
      "sourceUrl": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/",
      "verifiedDate": "2026-07-10",
      "status": "representative",
      "note": "Public preview, US-only with a waitlist at launch. Artificial Analysis scores it at Intelligence Index 43."
    },
    {
      "name": "Mistral",
      "slug": "mistral",
      "category": "first-party-lab",
      "region": "EU",
      "summary": "European (French) first-party lab; open-weight friendly with EU data residency.",
      "offering": "Mistral Large 3, Medium 3.5, Small 3, Codestral (code), Ministral 3B/8B (edge), Pixtral (vision), OCR. Many under Apache 2.0.",
      "pricingModel": "per-token",
      "representativePricing": "Large 2 tier $2/$6, Small 3 $0.10/$0.30, Ministral 3B ~$0.04/$0.04 per Mtok.",
      "standout": "Among the cheapest flagship-tier output pricing, plus EU data residency and genuine open weights for self-hosting.",
      "flagshipModel": "Mistral Large",
      "flagshipOutputPerMtok": 6,
      "openAICompatible": true,
      "freeTier": "Free experimentation tier via la Plateforme.",
      "homeUrl": "https://mistral.ai",
      "pricingUrl": "https://mistral.ai/pricing",
      "docsUrl": "https://console.mistral.ai",
      "sourceUrl": "https://mistral.ai/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "DeepSeek",
      "slug": "deepseek",
      "category": "first-party-lab",
      "region": "China",
      "summary": "Chinese first-party lab (open-weight); among the cheapest frontier-class APIs in the market.",
      "offering": "V4 Flash (cheapest frontier-class API, re-post-trained as the 0731 checkpoint on July 31, 2026) and V4 Pro. Both 1M ctx, 384K max output. Open weights with 10M+ downloads.",
      "pricingModel": "per-token",
      "representativePricing": "Peak rates: V4 Flash $0.44/$1.32, V4 Pro $1.32/$3.96 per Mtok, from DeepSeek's own price page. Off-peak is exactly half: $0.22/$0.66 and $0.66/$1.98. Cache hits are $0.014 and $0.044 at peak, still a roughly 97% discount on cache-miss input.",
      "standout": "Roughly 90-95% cheaper than comparable Western models, with open weights. Its cheap V4 Flash tier now outscores its own V4 Pro on the Artificial Analysis Intelligence Index, 50 to 44. Owned by hedge fund High-Flyer.",
      "flagshipModel": "DeepSeek V4 Pro",
      "flagshipOutputPerMtok": 3.96,
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://deepseek.com",
      "pricingUrl": "https://api-docs.deepseek.com/quick_start/pricing",
      "docsUrl": "https://api-docs.deepseek.com",
      "sourceUrl": "https://api-docs.deepseek.com/quick_start/pricing",
      "verifiedDate": "2026-08-21",
      "status": "verified",
      "note": "Prices re-verified against DeepSeek's official price page on 2026-08-21. The figures carried here are the PEAK rates from the peak/off-peak card that took effect 16:00 UTC on 2026-08-16; off-peak is exactly half. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, which DeepSeek states as 09:00-12:00 and 14:00-18:00 Beijing time, 7 of 24 hours. The pending surcharge previously recorded here as having no effective date is this change, and it landed as a larger rise than a flat 2x: the base rate moved too, so V4 Pro output went from $0.87 flat to $1.98 off-peak and $3.96 peak. Some hosts serve at aggressive quantization. Legacy R1/V3.2 aliases retired July 24, 2026."
    },
    {
      "name": "Cohere",
      "slug": "cohere",
      "category": "first-party-lab",
      "region": "US",
      "summary": "Enterprise-focused first-party lab specializing in RAG and retrieval.",
      "offering": "Command (generation: Command A/R+/R/R7B), Embed (vectors), Rerank (neural reranking). Strong on data sovereignty and on-prem.",
      "pricingModel": "per-token",
      "representativePricing": "Command R+ / Command A $2.50/$10, Command R $0.15/$0.60, Command R7B $0.0375/$0.15 per Mtok. Embed v3 $0.10/M.",
      "standout": "Best-in-class Embed plus Rerank stack for retrieval; strong on data sovereignty, VPC and on-prem deployment.",
      "flagshipModel": "Command A",
      "flagshipOutputPerMtok": 10,
      "openAICompatible": false,
      "freeTier": "Free trial key (1,000 calls/month, not for production).",
      "homeUrl": "https://cohere.com",
      "pricingUrl": "https://cohere.com/pricing",
      "docsUrl": "https://docs.cohere.com",
      "sourceUrl": "https://cohere.com/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "MiniMax",
      "slug": "minimax",
      "category": "first-party-lab",
      "region": "China",
      "summary": "Chinese first-party lab; multimodal across text, voice and Hailuo video.",
      "offering": "M-series: M2.7, M3, plus legacy abab6.5 and MiniMax-01 (1M ctx). A faster highspeed variant at 2x.",
      "pricingModel": "per-token",
      "representativePricing": "M2.7 $0.30/$1.20 per Mtok (official), cache reads $0.06/M.",
      "standout": "Frontier-class coding and agentic quality at ~5-10% of Claude Opus output pricing. M2.7 restricts commercial use (M2 was MIT).",
      "flagshipModel": "MiniMax M2.7",
      "flagshipOutputPerMtok": 1.2,
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://minimax.io",
      "pricingUrl": "https://platform.minimax.io/docs/guides/pricing-paygo",
      "docsUrl": "https://platform.minimax.io/docs",
      "sourceUrl": "https://platform.minimax.io/docs/guides/pricing-paygo",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "StepFun",
      "slug": "stepfun",
      "category": "first-party-lab",
      "region": "China",
      "summary": "Chinese first-party lab; multimodal MoE models, mostly open-weight.",
      "offering": "Step 3.7 Flash (196B/~11B active, 256K ctx, native image and video input), Step 3.5 Flash, Step3.",
      "pricingModel": "per-token",
      "representativePricing": "Step 3.7 Flash $0.20/$1.15, Step 3.5 Flash $0.09/$0.30, Step3 $0.57/$1.42 per Mtok.",
      "standout": "Disproportionately strong agentic benchmark scores relative to its price tier.",
      "flagshipModel": "Step3",
      "flagshipOutputPerMtok": 1.42,
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://stepfun.ai",
      "pricingUrl": "https://platform.stepfun.ai/docs/en/guides/pricing/details",
      "docsUrl": "https://platform.stepfun.ai/docs",
      "sourceUrl": "https://platform.stepfun.ai/docs/en/guides/pricing/details",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Reka AI",
      "slug": "reka-ai",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party natively-multimodal lab across text, image, video and audio.",
      "offering": "Reka Core (top reasoning), Flash/Flash 3 (21B), Edge (7B vision-language), Spark. Deployable cloud, on-prem or on-device.",
      "pricingModel": "per-token",
      "representativePricing": "Edge $0.10/$0.10, Flash 3 $0.10/$0.20 per Mtok (via OpenRouter); Core is most expensive.",
      "standout": "Flexible deployment down to edge and device; Reka Edge uses only 64 tokens per image tile for low-latency robotics and AR.",
      "freeTier": "",
      "homeUrl": "https://reka.ai",
      "pricingUrl": "https://docs.reka.ai/pricing",
      "docsUrl": "https://docs.reka.ai",
      "sourceUrl": "https://docs.reka.ai/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Pricing via OpenRouter; official rates may differ modestly."
    },
    {
      "name": "Upstage",
      "slug": "upstage",
      "category": "first-party-lab",
      "region": "Korea",
      "summary": "South Korean first-party lab; also document AI.",
      "offering": "Solar Pro 3 (102B total/12B active, 128K ctx, tuned for Korean/English/Japanese), Solar Pro 2, Document Parse/Extract.",
      "pricingModel": "per-token",
      "representativePricing": "Solar Pro 3 ~$0.15/$0.60 per Mtok (via OpenRouter); Document Parse ~$0.01/page. Prices exclude 10% VAT.",
      "standout": "Best positioned for Korean-language and structured document and instruction-following tasks.",
      "flagshipModel": "Solar Pro 3",
      "flagshipOutputPerMtok": 0.6,
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://upstage.ai",
      "pricingUrl": "https://upstage.ai/pricing/api",
      "docsUrl": "https://console.upstage.ai/docs",
      "sourceUrl": "https://upstage.ai/pricing/api",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Kimi (Moonshot AI)",
      "slug": "kimi-moonshot",
      "category": "first-party-lab",
      "region": "China",
      "summary": "Chinese first-party lab; long-context and agentic specialist, open-weight (modified MIT).",
      "offering": "Kimi K3 (flagship, ~2.8T-param MoE, 1M ctx, native vision, weights due 2026-07-27), K2.6 (1T MoE/32B active, 256K ctx), K2.7-Code (coding specialist). Agent Swarm up to 300 subagents.",
      "pricingModel": "per-token",
      "representativePricing": "K3 $3.00/$15.00 (cache-hit input $0.30), K2.7-Code $0.95/$4.00 per Mtok. Batch API 40% off.",
      "standout": "Flagship K3 lands at Artificial Analysis Intelligence Index 57 with API pricing below Opus 4.8 and GPT-5.5; the K2 line stays far cheaper.",
      "flagshipModel": "Kimi K3",
      "flagshipOutputPerMtok": 15,
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://moonshot.ai",
      "pricingUrl": "https://platform.kimi.ai/docs/pricing",
      "docsUrl": "https://platform.kimi.ai",
      "sourceUrl": "https://platform.kimi.ai/docs/pricing/chat-k3",
      "verifiedDate": "2026-07-17",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Sarvam (Sarvam AI)",
      "slug": "sarvam-ai",
      "category": "first-party-lab",
      "region": "India",
      "summary": "Indian first-party full-stack sovereign AI lab, IndiaAI Mission-backed.",
      "offering": "Sarvam-30B and Sarvam-105B (MoE, trained on Indian compute, 128K ctx, open-weight), Bulbul (TTS), Saaras (STT), translation (22 languages), OCR.",
      "pricingModel": "per-token",
      "representativePricing": "Chat completion ~Rs 4 input / Rs 16 output per Mtok (105B tier); STT Rs 45/hour. Free credits on signup.",
      "standout": "INR-denominated pricing avoids USD plus GST overhead; data hosted in India; best-in-class Indic-language and OCR performance.",
      "openAICompatible": true,
      "freeTier": "Free credits on signup.",
      "homeUrl": "https://sarvam.ai",
      "pricingUrl": "https://sarvam.ai/api-pricing",
      "docsUrl": "https://docs.sarvam.ai",
      "sourceUrl": "https://sarvam.ai/api-pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Raised a $234M first close of a $300M Series B at a $1.5B valuation (June 15, 2026), led by HCLTech."
    },
    {
      "name": "Inception (Inception Labs)",
      "slug": "inception-labs",
      "category": "first-party-lab",
      "region": "US",
      "summary": "First-party lab pioneering diffusion LLMs (dLLMs) that generate tokens in parallel via denoising.",
      "offering": "Mercury 2 (reasoning dLLM, 128K ctx, >1,000 tok/s on Blackwell), Mercury Coder, Mercury Edit 2. Default in Continue and Zed.",
      "pricingModel": "per-token",
      "representativePricing": "Mercury 2 $0.25/$0.75 per Mtok (cached input $0.025/M).",
      "standout": "First commercially available diffusion LLM, for ~5-10x faster, cheaper inference than one-token-at-a-time models.",
      "flagshipModel": "Mercury 2",
      "flagshipOutputPerMtok": 0.75,
      "openAICompatible": true,
      "freeTier": "10M free tokens per new account.",
      "homeUrl": "https://inceptionlabs.ai",
      "pricingUrl": "https://inceptionlabs.ai/models",
      "docsUrl": "https://inceptionlabs.ai/models",
      "sourceUrl": "https://inceptionlabs.ai/models",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Stanford spinout, ~$50M raised."
    },
    {
      "name": "Microsoft Azure",
      "slug": "azure",
      "category": "hyperscaler-marketplace",
      "region": "US",
      "summary": "Cloud hyperscaler with an 11,000+ model marketplace (Azure OpenAI / AI Foundry).",
      "offering": "Hosts OpenAI (GPT-5 family, Sora, image), plus DeepSeek, Grok, Llama, Mistral, FLUX, and managed GPU compute.",
      "pricingModel": "mixed",
      "representativePricing": "Token pricing matches OpenAI direct (GPT-5 $1.25/$10 per Mtok). PTUs for sustained load from ~$2,448/mo.",
      "standout": "Enterprise governance, compliance and Azure integration. Real bills often run 15-40% above raw token cost (support, networking, search).",
      "openAICompatible": true,
      "freeTier": "$200 free credit for 30 days.",
      "homeUrl": "https://azure.microsoft.com",
      "pricingUrl": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/",
      "docsUrl": "https://learn.microsoft.com/azure/ai-services/openai/",
      "sourceUrl": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Amazon Bedrock",
      "slug": "amazon-bedrock",
      "category": "hyperscaler-marketplace",
      "region": "US",
      "summary": "Cloud hyperscaler managed model service with a 100+ model marketplace.",
      "offering": "Foundation models from Anthropic, Meta, Mistral, Cohere, AI21, Amazon (Nova/Titan), Stability, and now OpenAI. AgentCore for agents.",
      "pricingModel": "mixed",
      "representativePricing": "Per-token, matching providers (Claude Sonnet 5 $3/$15, Nova Micro $0.035/$0.14 per Mtok). Batch 50% off; caching up to 90% off.",
      "standout": "Deep AWS integration, FedRAMP and HIPAA compliance. Watch hidden costs (OpenSearch Serverless ~$345/mo for Knowledge Bases).",
      "openAICompatible": false,
      "freeTier": "",
      "homeUrl": "https://aws.amazon.com/bedrock",
      "pricingUrl": "https://aws.amazon.com/bedrock/pricing",
      "docsUrl": "https://docs.aws.amazon.com/bedrock/",
      "sourceUrl": "https://aws.amazon.com/bedrock/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Five billing modes: on-demand, batch, provisioned throughput, prompt caching, customization."
    },
    {
      "name": "Snowflake (Cortex)",
      "slug": "snowflake-cortex",
      "category": "hyperscaler-marketplace",
      "region": "US",
      "summary": "AI inference layer inside the Snowflake Data Cloud; data never leaves the perimeter.",
      "offering": "Pre-integrated Arctic, Llama, Mistral, Reka, Google, plus OpenAI, Anthropic, DeepSeek. AISQL, Cortex Search, Analyst and Agents.",
      "pricingModel": "consumption",
      "representativePricing": "Consumption and credit-based, token-metered, roughly $0.12-5.10 per Mtok depending on model; warehouse compute billed separately.",
      "standout": "AI runs where the data lives, no egress; Snowflake does not train on customer data.",
      "openAICompatible": false,
      "freeTier": "No dedicated free tier (trial credits only).",
      "homeUrl": "https://www.snowflake.com/en/product/features/cortex/",
      "pricingUrl": "https://docs.snowflake.com/en/user-guide/snowflake-cortex",
      "docsUrl": "https://docs.snowflake.com/en/user-guide/snowflake-cortex",
      "sourceUrl": "https://docs.snowflake.com/en/user-guide/snowflake-cortex",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Databricks (Mosaic AI)",
      "slug": "databricks-mosaic-ai",
      "category": "hyperscaler-marketplace",
      "region": "US",
      "summary": "Data and AI platform with model serving on the lakehouse, under unified governance.",
      "offering": "Open foundation models (Llama, DBRX) plus external models (OpenAI, Anthropic, Cohere). Tight Unity Catalog governance.",
      "pricingModel": "consumption",
      "representativePricing": "Consumption via DBUs from ~$0.07/DBU; pay-per-token, provisioned throughput (Llama 3.3 70B from $6/hr per band), and batch.",
      "standout": "Tight integration with Unity Catalog governance and data pipelines; OpenAI-compatible API.",
      "openAICompatible": true,
      "freeTier": "14-day free trial; Free Edition available.",
      "homeUrl": "https://databricks.com",
      "pricingUrl": "https://databricks.com/product/pricing/foundation-model-serving",
      "docsUrl": "https://docs.databricks.com/en/machine-learning/model-serving/",
      "sourceUrl": "https://databricks.com/product/pricing/foundation-model-serving",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Together AI",
      "slug": "together-ai",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Neutral open-model inference cloud hosting 200+ open models.",
      "offering": "Serverless per-token (DeepSeek, Llama, Qwen, Kimi, GLM, MiniMax, Mixtral), plus fine-tuning, dedicated endpoints and GPU clusters.",
      "pricingModel": "mixed",
      "representativePricing": "DeepSeek V3.1 $0.60/$1.70, GPT-OSS 20B $0.05/$0.20 per Mtok; H100 reserved ~$3.99/hr. $5 minimum credit.",
      "standout": "Broad catalog, research-driven optimization (FlashAttention lineage), full-stack from serverless to clusters.",
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://together.ai",
      "pricingUrl": "https://together.ai/pricing",
      "docsUrl": "https://docs.together.ai",
      "sourceUrl": "https://together.ai/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Fireworks AI",
      "slug": "fireworks-ai",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Neutral open-model inference platform from the creators of PyTorch; speed and throughput focus.",
      "offering": "Serverless per-token (DeepSeek/Kimi/GLM/MiniMax catalog), on-demand GPUs, fine-tuning, reserved capacity. 30T+ tokens/day.",
      "pricingModel": "mixed",
      "representativePricing": "8B-class ~$0.20/M, 70B-class ~$0.90/M; H100/H200 $6/hr, B200 $9/hr. Batch 50% off. Often 20-40% below Together.",
      "standout": "Day-zero model support and rapid growth: $315M annualized revenue as of February 2026; in talks at a $15B valuation in mid-2026.",
      "openAICompatible": true,
      "freeTier": "$1 free starter credit.",
      "homeUrl": "https://fireworks.ai",
      "pricingUrl": "https://fireworks.ai/pricing",
      "docsUrl": "https://docs.fireworks.ai",
      "sourceUrl": "https://fireworks.ai/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "DeepInfra",
      "slug": "deepinfra",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Low-cost serverless inference for open-weight models; runs its own US data centers.",
      "offering": "190+ models (Llama, Qwen, DeepSeek, GLM, Gemma, Mistral, Nemotron, Kimi) plus embeddings, TTS, image. Dedicated GPUs by the hour.",
      "pricingModel": "mixed",
      "representativePricing": "From ~$0.06/M for small models; DeepSeek V4 Flash $0.10/$0.20 per Mtok. ~5T tokens/week.",
      "standout": "Among the cheapest serverless options, runs its own US data centers including Blackwell B200.",
      "openAICompatible": true,
      "freeTier": "No standing free tier.",
      "homeUrl": "https://deepinfra.com",
      "pricingUrl": "https://deepinfra.com/pricing",
      "docsUrl": "https://deepinfra.com/docs",
      "sourceUrl": "https://deepinfra.com/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Community reports flag FP4 quantization on some models."
    },
    {
      "name": "Replicate",
      "slug": "replicate",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Developer-friendly model-hosting marketplace (acquired by Cloudflare, late 2025).",
      "offering": "50,000+ community models plus ~100 curated Official Models (Claude, DeepSeek, FLUX, Veo, Kling). Deploy custom models via Cog.",
      "pricingModel": "mixed",
      "representativePricing": "Hardware-per-second (CPU $0.000025/s up to H100 ~$0.001525/s) and output-based (per token/image/video).",
      "standout": "Largest open model catalog and easiest experimentation. Cold starts and unpredictable per-second billing are the main drawbacks.",
      "openAICompatible": false,
      "freeTier": "",
      "homeUrl": "https://replicate.com",
      "pricingUrl": "https://replicate.com/pricing",
      "docsUrl": "https://replicate.com/docs",
      "sourceUrl": "https://replicate.com/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "Now owned by Cloudflare, which may affect the roadmap over time."
    },
    {
      "name": "SiliconFlow",
      "slug": "siliconflow",
      "category": "neutral-inference-platform",
      "region": "China",
      "summary": "Chinese model-as-a-service aggregator; notable for Huawei Ascend-based DeepSeek serving.",
      "offering": "200+ models across text/image/video/audio (DeepSeek, Qwen, GLM, Kimi, MiniMax, Step, FLUX). Backed by Alibaba Cloud.",
      "pricingModel": "mixed",
      "representativePricing": "Pay-as-you-go per-token (DeepSeek V4 Flash $0.14/$0.28 per Mtok); reserved GPU ~CNY 2.73/hr.",
      "standout": "6M+ users, 100B+ daily tokens, fastest to handle DeepSeek traffic.",
      "openAICompatible": true,
      "freeTier": "$1 credits on signup; some smaller models permanently free.",
      "homeUrl": "https://siliconflow.com",
      "pricingUrl": "https://siliconflow.com/pricing",
      "docsUrl": "https://docs.siliconflow.com",
      "sourceUrl": "https://siliconflow.com/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Hyperbolic",
      "slug": "hyperbolic",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Decentralized open-access AI cloud: a GPU marketplace plus serverless inference.",
      "offering": "25+ open models plus a GPU marketplace that aggregates idle GPUs. Pay by card or crypto. Serves Llama-3.1-405B-Base in BF16.",
      "pricingModel": "mixed",
      "representativePricing": "GPU/hr: RTX 4090 $0.50, A100 ~$1.60-1.80, H100 PCIe $3.00, H100 SXM $3.20. Llama 3.3 70B $0.40/M.",
      "standout": "Up to 75% savings vs hyperscalers by pooling idle GPUs; only platform serving Llama-3.1-405B-Base in BF16.",
      "openAICompatible": true,
      "freeTier": "$1 promo credit (not for GPU rental).",
      "homeUrl": "https://hyperbolic.ai",
      "pricingUrl": "https://docs.hyperbolic.xyz/docs/hyperbolic-pricing",
      "docsUrl": "https://docs.hyperbolic.xyz",
      "sourceUrl": "https://docs.hyperbolic.xyz/docs/hyperbolic-pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Nebius",
      "slug": "nebius",
      "category": "neutral-inference-platform",
      "region": "EU",
      "summary": "Full-stack European AI cloud (successor to Yandex N.V.); Token Factory plus raw GPU rental.",
      "offering": "Token Factory: 60+ open-source models in Fast/Base tiers, OpenAI-compatible. Also H100/H200/B200/B300 rental with per-second billing.",
      "pricingModel": "mixed",
      "representativePricing": "Cheapest ~$0.06-0.08/M input (Nemotron 3 Nano $0.08 blended) up to ~$1.93/M for DeepSeek V4 Pro. Reserve discounts up to 35%.",
      "standout": "EU data residency, full-stack managed compute plus inference, and a ~$2B NVIDIA deal with early Rubin access.",
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://nebius.com",
      "pricingUrl": "https://nebius.com/token-factory/prices",
      "docsUrl": "https://docs.nebius.com",
      "sourceUrl": "https://nebius.com/token-factory/prices",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "The NVIDIA Rubin 10x cost-per-token claim is projected for H2 2026, not yet delivered."
    },
    {
      "name": "Parasail",
      "slug": "parasail",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Inference-only AI Supercloud aggregating global GPU supply (40 data centers, 15 countries).",
      "offering": "Serverless per-token (GLM-5.2, Kimi, DeepSeek V4, MiniMax, Qwen, gpt-oss, Llama), dedicated endpoints and batch. Deploy any HF model in ~5 lines.",
      "pricingModel": "mixed",
      "representativePricing": "~$0.09/M for MiMo-V2.5 up to ~$0.90/M for GLM-5.1. 500B tokens/day.",
      "standout": "Lossless by default, no hidden quantization; pay-per-token with no GPU contracts. Founded by an ex-Groq exec; raised $32M Series A.",
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://parasail.io",
      "pricingUrl": "https://saas.parasail.io/pricing",
      "docsUrl": "https://docs.parasail.io",
      "sourceUrl": "https://saas.parasail.io/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "FriendliAI",
      "slug": "friendliai",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Fast-serving inference platform with custom kernels and speculative decoding.",
      "offering": "Serverless Endpoints (OpenAI-compatible), Dedicated Endpoints (per GPU-hour), Container (on-prem). DeepSeek, Qwen, Kimi, GLM, Llama, EXAONE.",
      "pricingModel": "mixed",
      "representativePricing": "Pay-per-token serverless plus per GPU-hour dedicated. Claims 50-90% cost savings vs vLLM.",
      "standout": "Claims up to 3x faster than vLLM via custom kernels, speculative decoding, continuous batching. SOC 2 Type II plus HIPAA, 99.99% uptime SLA.",
      "openAICompatible": true,
      "freeTier": "$5 free credits.",
      "homeUrl": "https://friendli.ai",
      "pricingUrl": "https://friendli.ai/pricing",
      "docsUrl": "https://friendli.ai/docs",
      "sourceUrl": "https://friendli.ai/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Baseten",
      "slug": "baseten",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "Enterprise model-inference platform; deploy custom/open/fine-tuned models via the Truss framework.",
      "offering": "Per-minute dedicated GPU billing (T4 ~$0.01/min up to B200 ~$0.166/min) with scale-to-zero, plus a per-token Model APIs catalog.",
      "pricingModel": "mixed",
      "representativePricing": "Model APIs median ~$0.60/$2.20 per Mtok; dedicated GPU T4 ~$0.01/min up to B200 ~$0.166/min.",
      "standout": "Multi-cloud routing across ~18-20 providers, 1B+ inference calls/day. Closed a $1.5B Series F at a $13B valuation (June 22, 2026).",
      "openAICompatible": true,
      "freeTier": "New-account credits.",
      "homeUrl": "https://baseten.co",
      "pricingUrl": "https://baseten.co/pricing",
      "docsUrl": "https://docs.baseten.co",
      "sourceUrl": "https://baseten.co/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "GMI (GMI Cloud)",
      "slug": "gmi-cloud",
      "category": "neutral-inference-platform",
      "region": "US",
      "summary": "AI-native GPU and inference cloud (NVIDIA Preferred Partner); bare-metal pricing with no forced upsell.",
      "offering": "Bare-metal plus serverless plus dedicated clusters. Inference Engine (auto-scaling), Cluster Engine, and Model-as-a-Service.",
      "pricingModel": "per-gpu-hour",
      "representativePricing": "On-demand GPU/hr: H100 from $2.00, H200 from $2.60, B200 from $4.00, GB200 from $8.00. Reserved cuts 30-50%.",
      "standout": "Claims 40-70% savings vs hyperscalers; pure bare-metal rates with no forced CPU or networking upsell.",
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://gmicloud.ai",
      "pricingUrl": "https://gmicloud.ai/en/pricing",
      "docsUrl": "https://docs.gmicloud.ai",
      "sourceUrl": "https://gmicloud.ai/en/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "Groq",
      "slug": "groq",
      "category": "custom-silicon",
      "region": "US",
      "summary": "Custom-silicon (LPU) inference cloud for open models; among the fastest inference available.",
      "offering": "Runs Llama, Qwen, Kimi, GPT-OSS, DeepSeek distills and Whisper at 500-1,000+ tok/s. Catalog is open-source only.",
      "pricingModel": "per-token",
      "representativePricing": "Llama 3.1 8B $0.05/$0.08, Llama 3.3 70B $0.59/$0.79, Kimi K2 $1/$3 per Mtok. Batch and caching each cut 50%.",
      "standout": "Among the fastest inference available. NVIDIA agreed to pay ~$20B for a perpetual license to Groq's LPU patents (finalized December 24, 2025); GroqCloud continues operating.",
      "openAICompatible": true,
      "freeTier": "Free developer tier (no credit card).",
      "homeUrl": "https://groq.com",
      "pricingUrl": "https://groq.com/pricing",
      "docsUrl": "https://console.groq.com/docs",
      "sourceUrl": "https://groq.com/pricing",
      "verifiedDate": "2026-09-01",
      "status": "representative",
      "note": "",
      "detail": {
        "metaTitle": "Groq Pricing 2026: LPU Speed at Open-Model Rates",
        "metaDescription": "What Groq costs per million tokens in 2026, how its LPU speed premium compares against the same open models hosted elsewhere, and where the free plan stops.",
        "intro": "Groq sells inference on its own LPU silicon rather than on GPUs, and the whole proposition is latency: 500 to 1,000-plus tokens per second on a catalog that is deliberately open-weight only. It is a speed vendor first and a price vendor second, and on the models where those two things can be compared directly, the pricing is more interesting than the marketing suggests.",
        "whatItServes": "The catalog is open-source models only: Llama, Qwen, Kimi, GPT-OSS and DeepSeek distills for text, plus Whisper for speech. There is no proprietary frontier model here, and no facility to upload your own weights. Groq is a place to run models someone else trained, very fast, from a US region.",
        "howPricingWorks": "Billing is per token with no subscription, and the rate depends entirely on which model you call: roughly $0.05 input and $0.08 output per million tokens for Llama 3.1 8B, up to $1 and $3 for Kimi K2. Batch submission and prompt caching each cut the rate by half and stack, so a cached batch workload pays about a quarter of the on-demand price. The number worth knowing is comparative. On GPT-OSS-120B, Groq lists $0.15 input and $0.60 output per million tokens, which is exactly what Together, Nebius, Amazon Bedrock and DeepInfra charge for the identical weights, with cached input reads at $0.075. The speed is not billed as a premium on that row: it is the common market rate with a faster machine behind it.",
        "limitations": "Free-plan limits are set per model rather than per account, and they are tight: several models cap at 30 requests a minute, 1,000 a day, 8,000 tokens a minute and 200,000 tokens a day, and the documentation says the authoritative figures are the ones in your own account settings rather than the published table. Higher limits are described as available for select workloads and enterprise use. The open-weight-only catalog is the hard ceiling: no frontier proprietary model, no custom weights. One structural item belongs on any Groq decision: NVIDIA agreed to pay about $20 billion for a perpetual license to Groq LPU patents, finalized on December 24 2025. GroqCloud continues to operate, but a platform whose core silicon IP has been licensed to the dominant GPU vendor carries a roadmap question that a pricing page does not answer.",
        "whoItIsFor": "Latency-bound work on open models: voice agents, interactive coding assistants, search and rerank paths, anything where a person is waiting on the first token. It suits a team that has already chosen an open-weight model and wants it served faster than a GPU host will serve it. It does not suit a team that needs a frontier proprietary model, custom fine-tuned weights, or one vendor for both.",
        "relatedSlugs": [
          "cerebras",
          "sambanova"
        ],
        "faqs": [
          {
            "q": "How much does Groq cost per million tokens?",
            "a": "It depends on the model, and there is no subscription. Rates run from about $0.05 input and $0.08 output per million tokens on Llama 3.1 8B to $1 and $3 on Kimi K2, with GPT-OSS-120B at $0.15 and $0.60. Batch submission and prompt caching each halve the rate and stack, so a cached batch job pays roughly a quarter of on-demand."
          },
          {
            "q": "Is Groq cheaper than other inference providers?",
            "a": "Not systematically, no. On GPT-OSS-120B its $0.15 input rate matches Together, Nebius, Amazon Bedrock and DeepInfra to the cent for identical weights, while FlexAI lists the same model at $0.039. Groq competes on tokens per second, not on being the cheapest place to run a given model."
          },
          {
            "q": "Does Groq have a free tier?",
            "a": "Yes, a free developer plan with no credit card required. The limits are applied per model and are small, on the order of 30 requests a minute and 200,000 tokens a day on several models, so it is sized for evaluation rather than for serving traffic."
          },
          {
            "q": "Can you run GPT-5 or Claude on Groq?",
            "a": "No. The catalog is open-weight models only, so Llama, Qwen, Kimi, GPT-OSS and DeepSeek distills are available and the proprietary frontier models from OpenAI, Anthropic and Google are not. GPT-OSS is OpenAI open-weight release, not the GPT-5 family."
          }
        ]
      }
    },
    {
      "name": "Cerebras",
      "slug": "cerebras",
      "category": "custom-silicon",
      "region": "US",
      "summary": "Wafer-scale (WSE-3) inference cloud; the speed champion at 1,800-3,000+ tok/s on open models.",
      "offering": "Llama, Qwen, DeepSeek distills, GPT-OSS. No proprietary frontier models and no custom-model uploads.",
      "pricingModel": "per-token",
      "representativePricing": "~$0.10-6 per Mtok depending on model ($0.35/M cheapest input). Pay-as-you-go and enterprise tiers.",
      "standout": "Fastest inference benchmarked by Artificial Analysis.",
      "openAICompatible": true,
      "freeTier": "Free trial: $5 in credits, all models.",
      "homeUrl": "https://cerebras.ai",
      "pricingUrl": "https://cerebras.ai/pricing",
      "docsUrl": "https://inference-docs.cerebras.ai",
      "sourceUrl": "https://cerebras.ai/pricing",
      "verifiedDate": "2026-09-01",
      "status": "representative",
      "note": "",
      "detail": {
        "metaTitle": "Cerebras Pricing 2026: What Wafer-Scale Costs",
        "metaDescription": "What Cerebras charges per million tokens in 2026, the wafer-scale speed premium measured against the same model elsewhere, and what the free tier really is.",
        "intro": "Cerebras serves inference from wafer-scale WSE-3 parts and is repeatedly benchmarked as the fastest hosted inference available, at 1,800 to 3,000-plus tokens per second on open models. The company frames this as up to 30 times faster than GPU systems. The question a buyer needs answered is what that speed costs relative to the identical model elsewhere, and that number is knowable.",
        "whatItServes": "An open-weight catalog: Llama, Qwen including Qwen3 235B and Qwen3 Coder 480B, DeepSeek distills and GPT-OSS. There are no proprietary frontier models, and custom model weights and fine-tuning are enterprise-tier features rather than self-serve ones. Serving is from US infrastructure.",
        "howPricingWorks": "Three tiers. A free trial gives $5 in credits after account creation with access to all models. A self-serve Developer tier starts at $10 and carries ten times the free rate limits plus elevated processing priority. Enterprise adds the highest limits, custom weight support, fine-tuning and dedicated support. Per-token rates run roughly $0.10 to $6 per million tokens depending on model. The comparison that matters: on GPT-OSS-120B, Cerebras lists $0.35 input and $0.75 output per million tokens, against $0.15 and $0.60 at Groq, Together, Nebius, Amazon Bedrock and DeepInfra for the identical weights, and $0.039 and $0.10 at FlexAI. So Cerebras is about 2.3 times the common input rate and roughly 9 times the cheapest listing, for the same model. That is the wafer-scale premium stated as a number rather than as a claim.",
        "limitations": "The free tier is $5 in credits, not the 1,000,000 tokens per day figure that circulates widely in secondary coverage of Cerebras. Check the pricing page rather than a roundup before budgeting against it. The catalog is open-weight only, so there is no path to a frontier proprietary model. Custom weights and fine-tuning sit behind the enterprise tier, which makes Cerebras a poor fit for a team whose differentiation is its own tuned model unless it is willing to have that conversation. And the premium is real: on any workload where throughput rather than first-token latency is the binding constraint, the same model is available for well under half the price elsewhere.",
        "whoItIsFor": "Workloads where latency is the product and the model is already an open-weight one: real-time voice, agentic loops that chain many short calls, and interactive tools where a two-second wait loses the user. It is a poor fit for batch or offline work, where nobody is waiting and the 2.3-times input premium over the same weights elsewhere buys nothing, and for teams that need custom fine-tuned weights on a self-serve plan.",
        "relatedSlugs": [
          "groq",
          "sambanova"
        ],
        "faqs": [
          {
            "q": "How much does Cerebras inference cost?",
            "a": "Roughly $0.10 to $6 per million tokens depending on the model, billed per token. As a concrete anchor, GPT-OSS-120B is listed at $0.35 input and $0.75 output per million tokens. A self-serve Developer tier starts at $10 and an enterprise tier is quoted separately."
          },
          {
            "q": "Is Cerebras more expensive than Groq?",
            "a": "On the same model, yes. GPT-OSS-120B is $0.35 input and $0.75 output per million tokens on Cerebras against $0.15 and $0.60 on Groq, so about 2.3 times the input rate for identical weights. Cerebras is generally the faster of the two, so the premium buys tokens per second, not capability."
          },
          {
            "q": "Does Cerebras have a free tier?",
            "a": "It offers a free trial of $5 in credits after you create an account, with access to all models and community Discord support. Secondary coverage often reports a 1,000,000 token per day free allowance, which does not match the current pricing page."
          },
          {
            "q": "Can you fine-tune a model on Cerebras?",
            "a": "Not on the self-serve tiers. Custom model weight support and fine-tuning services are listed as enterprise-tier features, alongside the highest rate limits and a dedicated support team. The free and Developer tiers serve the stock open-weight catalog only."
          }
        ]
      }
    },
    {
      "name": "SambaNova",
      "slug": "sambanova",
      "category": "custom-silicon",
      "region": "US",
      "summary": "Custom-silicon (RDU) inference platform (SambaCloud) purpose-built for agentic inference.",
      "offering": "Fast serving of large open models (Llama, DeepSeek 671B, Qwen, MiniMax) with a three-tier memory architecture.",
      "pricingModel": "per-token",
      "representativePricing": "Pay-per-token; rates listed on the SambaCloud plans page.",
      "standout": "Benchmarked by Artificial Analysis as among the fastest for large models (MiniMax M2.7 at 435 tok/s). Strong sovereign-AI and on-prem story.",
      "openAICompatible": true,
      "freeTier": "",
      "homeUrl": "https://sambanova.ai",
      "pricingUrl": "https://cloud.sambanova.ai/plans",
      "docsUrl": "https://docs.sambanova.ai",
      "sourceUrl": "https://cloud.sambanova.ai/plans",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "CoreWeave",
      "slug": "coreweave",
      "category": "gpu-compression-niche",
      "region": "US",
      "summary": "Specialized GPU hyperscaler (neocloud), Kubernetes-native, for AI training and inference at scale.",
      "offering": "Rents NVIDIA A100, H100, H200, GB200/B200, GB300. Per-second billing, no egress fees; spot and reserved available.",
      "pricingModel": "per-gpu-hour",
      "representativePricing": "8x H100 node ~$49.24/hr (~$6.16/GPU/hr); single GPUs from ~$1.19/hr (A100 PCIe) to $10.50/hr (B200 NVL).",
      "standout": "Roughly 40-60% cheaper than hyperscalers for equivalent GPUs; customers include OpenAI, Mistral, Jane Street. Often 8-GPU minimums.",
      "openAICompatible": false,
      "freeTier": "",
      "homeUrl": "https://coreweave.com",
      "pricingUrl": "https://coreweave.com/pricing",
      "docsUrl": "https://docs.coreweave.com",
      "sourceUrl": "https://coreweave.com/pricing",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": ""
    },
    {
      "name": "CompactifAI (Multiverse Computing)",
      "slug": "compactifai",
      "category": "gpu-compression-niche",
      "region": "Spain",
      "summary": "Quantum-inspired model-compression company that shrinks open LLMs up to 95% via tensor networks.",
      "offering": "Serves compressed Slim models via API and the AWS/Azure marketplaces. HyperNova 60B (from gpt-oss-120b), compressed Llama/DeepSeek/Mistral.",
      "pricingModel": "per-token",
      "representativePricing": "HyperNova 60B $0.04/$0.14 per Mtok; Llama 3.3 70B Slim ~$0.15/$0.31 per Mtok.",
      "standout": "Claims compressed models beat their base models on speed and cost, runnable on edge devices down to Raspberry Pi. Raised a 189M euro Series B (June 12, 2025).",
      "openAICompatible": false,
      "freeTier": "",
      "homeUrl": "https://multiversecomputing.com/compactifai",
      "pricingUrl": "https://multiversecomputing.com/compactifai/api",
      "docsUrl": "https://multiversecomputing.com/compactifai/api",
      "sourceUrl": "https://multiversecomputing.com/compactifai/api",
      "verifiedDate": "2026-06-28",
      "status": "representative",
      "note": "States up to 95% compression with only 2-3% precision loss (vendor claim)."
    }
  ]
}