{"asOf":"2026-09-04","methodology":"value = benchmark composite score / blended price; blended price = (3 * input + output) / 4 in USD per million tokens. Benchmark scores are composites (0-100) from the source dataset; prices are verified per-token API rates.","benchmarkSource":{"name":"Price Per Token","url":"https://pricepertoken.com/","note":"Composite Coding and Intelligence scores read from the Price Per Token dataset, which aggregates independent benchmarks (it cites Artificial Analysis, the HuggingFace Open LLM Leaderboard, and LayerLens). Scores are composites on a 0-100 scale, not a single named test. Prices marked verified are reconciled with this site's own registry (ai-models.json) against the provider source; unmarked prices are the author-direct or primary rate reported by Price Per Token (its pricing_source noted per row) and are not yet independently re-verified. Scores for the newest frontier models that Price Per Token has not published a composite for yet (GPT-5.6 Sol, Terra, Luna, Grok 4.5, Thinking Machines Inkling, Kimi K3, and both DeepSeek V4 tiers) are read directly from Artificial Analysis, the same independent benchmark family Price Per Token aggregates, and are flagged per row (scoreSource). The DeepSeek V4 rows were re-read from Artificial Analysis on 2026-07-31: V4 Pro was corrected from 52 to 44. Both DeepSeek rows use the Reasoning / Max Effort configuration, so they are on the same basis as each other; the old 52 came from April launch reporting for a differently labelled configuration. Price Per Token's own stale intelligence figure for V4 Pro (31.2) is not used. The Artificial Analysis Coding Agent Index and this board's coding composite are close but not identical scales, so treat cross-source coding ranks as approximate. INDEX DRIFT, 2026-08-07: every Artificial-Analysis-sourced Intelligence Index value on this board was re-read on one day, because the index had drifted upward by 1 to 3 points across the board since the mid-July reads (Kimi K3 57 to 60, GPT-5.6 Sol 59 to 61, Terra 55 to 57, Luna 51 to 52, Grok 4.5 54 to 56, Inkling 41 to 42, DeepSeek V4 Flash 50 to 52, V4 Pro 44 to 45, and in the unscored notes Claude Opus 5 61 to 63 and Claude Sonnet 5 53 to 55). No model was re-released in that window, so the movement is recalibration or endpoint re-testing on Artificial Analysis's side rather than model change. The practical rule this establishes: Intelligence Index figures read on different dates cannot be compared against each other, and a same-day re-read is required before any row is added. Effort-variant labels are now recorded per row for the same reason. Only Intelligence Index values were re-read on 2026-08-07; the Coding Agent Index figures are not exposed on the public models leaderboard and still carry their 2026-07-13 reading. PARTIAL RE-READ, 2026-09-04: a single-pass re-read of the Price Per Token composites found the coding figures stable for every carried-over model (Claude Opus 4.8 74.3, Claude Fable 5 76.5, Gemini 3.1 Pro 68.8, Qwen3.7 Max 66.0, Qwen3.7 Plus 55.9, Kimi K2.7 Code 60.8, MiniMax M3 58.6, Nemotron 3 Ultra 49.3) with ONE exception: GLM-5.2 fell from 68.8 to 46.5 on coding and 51.1 to 34.8 on intelligence. That row has been updated and re-dated; the rest are left at their prior reading. Intelligence values across the board again read 1 to 2 points above their 2026-08-07 figures (for example Opus 4.8 55.7 to 57.3, Fable 5 59.9 to 62.1, Kimi K2.7 Code 41.9 to 43.0), consistent with the documented drift, and are NOT updated here because a partial intelligence refresh would mix read dates on one axis. A full same-day re-read of every row is the outstanding task.","scoresVerified":"2026-07-13"},"rows":[{"provider":"Anthropic","model":"Claude Opus 4.8","origin":"United States","available":true,"inputPerMtok":5,"outputPerMtok":25,"blended":10,"coding":74.3,"intelligence":55.7,"codingValue":7.4,"intelligenceValue":5.6,"priceSource":"Anthropic (verified)","priceVerified":true,"priceSourceUrl":"https://claude.com/pricing","scoreSource":"","note":"Anthropic's prior flagship, superseded by Claude Opus 5 on July 24, 2026 at the same $5/$25 rate; still buyable on the API. Among the highest coding composites of any model here."},{"provider":"Anthropic","model":"Claude Sonnet 4.6","origin":"United States","available":true,"inputPerMtok":3,"outputPerMtok":15,"blended":6,"coding":null,"intelligence":34.3,"codingValue":null,"intelligenceValue":5.7,"priceSource":"Anthropic (verified)","priceVerified":true,"priceSourceUrl":"https://claude.com/pricing","scoreSource":"","note":"No standalone coding composite in the source dataset; intelligence composite only."},{"provider":"Anthropic","model":"Claude Haiku 4.5","origin":"United States","available":true,"inputPerMtok":1,"outputPerMtok":5,"blended":2,"coding":null,"intelligence":23.7,"codingValue":null,"intelligenceValue":11.9,"priceSource":"Anthropic (verified)","priceVerified":true,"priceSourceUrl":"https://claude.com/pricing","scoreSource":"","note":"Cheapest Claude tier. Intelligence composite only."},{"provider":"Anthropic","model":"Claude Fable 5","origin":"United States","available":true,"inputPerMtok":10,"outputPerMtok":50,"blended":20,"coding":76.5,"intelligence":59.9,"codingValue":3.8,"intelligenceValue":3,"priceSource":"Anthropic (verified)","priceVerified":true,"priceSourceUrl":"https://claude.com/pricing","scoreSource":"","note":"Anthropic's most capable model and the highest raw scores in the set. Suspended worldwide on June 12, 2026 under a US export-control directive, then restored globally on July 1, 2026, so it is buyable again and eligible for value picks. Its high token price keeps its value rank low despite top scores."},{"provider":"OpenAI","model":"GPT-5.6 Sol","origin":"United States","available":true,"inputPerMtok":4,"outputPerMtok":20,"blended":8,"coding":80,"intelligence":61,"codingValue":10,"intelligenceValue":7.6,"priceSource":"OpenAI (verified)","priceVerified":true,"priceSourceUrl":"https://developers.openai.com/api/docs/pricing","scoreSource":"Artificial Analysis","note":"Flagship of OpenAI's GPT-5.6 family (Sol, Terra, Luna), generally available July 9, 2026. Highest coding composite in the buyable set. Repriced 2026-08-21 to $4/$20 (from $5/$30), OpenAI's first Sol cut after excluding it from the July 30 Terra/Luna reductions; the new rate is promotional through at least 2026-11-21 and was verified on OpenAI's pricing page 2026-08-25. At the blended $12 rate Sol now undercuts Claude Opus 5 ($5/$25, blended $15) on price at the same AA Intelligence Index of 61, which materially improves this row's value rank. Scored from Artificial Analysis: Intelligence Index 61 at the Max Effort variant, re-read 2026-08-07 (up from the 59 this board carried, part of an index-wide upward drift, see the source note above). Coding Agent Index 80 was not re-read on 2026-08-07 and still carries its 2026-07-13 reading. Pending a Price Per Token composite."},{"provider":"OpenAI","model":"GPT-5.6 Terra","origin":"United States","available":true,"inputPerMtok":2,"outputPerMtok":12,"blended":4.5,"coding":77,"intelligence":57,"codingValue":17.1,"intelligenceValue":12.7,"priceSource":"OpenAI (verified)","priceVerified":true,"priceSourceUrl":"https://developers.openai.com/api/docs/pricing","scoreSource":"Artificial Analysis","note":"Mid tier of the GPT-5.6 family, repriced 20% lower on July 30, 2026 to $2/$12. That was the smaller of the two cuts in the same announcement, and it left Terra awkward: Luna fell 80% to $0.20/$1.20, so Terra now costs 10x Luna for 2 extra coding points. Scored from Artificial Analysis: Intelligence Index 57 at the Max Effort variant, re-read 2026-08-07 (up from 55). Coding Agent Index 77 still carries its 2026-07-13 reading. Pending a Price Per Token composite."},{"provider":"OpenAI","model":"GPT-5.6 Luna","origin":"United States","available":true,"inputPerMtok":0.2,"outputPerMtok":1.2,"blended":0.5,"coding":75,"intelligence":52,"codingValue":166.7,"intelligenceValue":115.6,"priceSource":"OpenAI (verified)","priceVerified":true,"priceSourceUrl":"https://developers.openai.com/api/docs/pricing","scoreSource":"Artificial Analysis","note":"Budget tier of the GPT-5.6 family, repriced 80% lower on July 30, 2026 to $0.20/$1.20. That cut makes it the best-value model on this board outright, not just among US models: a coding composite of 75 at a blended $0.45 per Mtok beats MiniMax M3, the previous value leader, on both halves of the ratio (58.6 at a blended $0.53). Scored from Artificial Analysis: Intelligence Index 52 at the Max Effort variant, re-read 2026-08-07 (up from 51). Coding Agent Index 75 still carries its 2026-07-13 reading. Pending a Price Per Token composite."},{"provider":"OpenAI","model":"GPT-5.3 Codex","origin":"United States","available":true,"inputPerMtok":1.75,"outputPerMtok":14,"blended":4.8,"coding":null,"intelligence":44.3,"codingValue":null,"intelligenceValue":9.2,"priceSource":"OpenAI (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://openai.com/api/pricing/","scoreSource":"","note":"OpenAI's prior agentic coding tier. Intelligence composite only; no coding composite recorded yet."},{"provider":"OpenAI","model":"GPT-5.2 Pro","origin":"United States","available":true,"inputPerMtok":21,"outputPerMtok":168,"blended":57.8,"coding":null,"intelligence":42.2,"codingValue":null,"intelligenceValue":0.7,"priceSource":"OpenAI (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://openai.com/api/pricing/","scoreSource":"","note":"The most expensive model in the set at $21/$168 per Mtok, which makes it the worst value despite a top-tier score. A clean illustration that price, not capability, drives the value ranking."},{"provider":"Google","model":"Gemini 3.1 Pro","origin":"United States","available":true,"inputPerMtok":2,"outputPerMtok":12,"blended":4.5,"coding":68.8,"intelligence":46.5,"codingValue":15.3,"intelligenceValue":10.3,"priceSource":"Google (verified)","priceVerified":true,"priceSourceUrl":"https://ai.google.dev/gemini-api/docs/pricing","scoreSource":"","note":"Flagship Pro tier, standard context. Source dataset labels it Preview; Google still markets it as a preview."},{"provider":"Google","model":"Gemini 3 Pro Preview","origin":"United States","available":true,"inputPerMtok":2,"outputPerMtok":12,"blended":4.5,"coding":null,"intelligence":33.1,"codingValue":null,"intelligenceValue":7.4,"priceSource":"Google (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://ai.google.dev/gemini-api/docs/pricing","scoreSource":"","note":"The prior Gemini 3 Pro generation, same list price as 3.1 Pro but a lower intelligence composite. Intelligence only."},{"provider":"Google","model":"Gemini 3.5 Flash","origin":"United States","available":true,"inputPerMtok":1.5,"outputPerMtok":9,"blended":3.4,"coding":null,"intelligence":34.9,"codingValue":null,"intelligenceValue":10.3,"priceSource":"Google (verified)","priceVerified":true,"priceSourceUrl":"https://ai.google.dev/gemini-api/docs/pricing","scoreSource":"","note":"The 3.5-generation Flash tier. Intelligence composite only."},{"provider":"xAI","model":"Grok 4.5","origin":"United States","available":true,"inputPerMtok":2,"outputPerMtok":6,"blended":3,"coding":76,"intelligence":56,"codingValue":25.3,"intelligenceValue":18.7,"priceSource":"xAI (verified)","priceVerified":true,"priceSourceUrl":"https://docs.x.ai/docs/models","scoreSource":"Artificial Analysis","note":"xAI's coding-and-agent model, released July 8, 2026 at $2/$6 per Mtok with a 500K context. Trained jointly with Cursor on coding-agent traces. Scored from Artificial Analysis: Intelligence Index 56 at the High Effort variant, the only Grok 4.5 configuration the leaderboard publishes, re-read 2026-08-07 (up from 54). Coding Agent Index 76 still carries its 2026-07-13 reading. Pending a Price Per Token composite."},{"provider":"Thinking Machines","model":"Inkling","origin":"United States","available":true,"inputPerMtok":1.87,"outputPerMtok":4.68,"blended":2.6,"coding":null,"intelligence":42,"codingValue":null,"intelligenceValue":16.3,"priceSource":"Thinking Machines Tinker (verified)","priceVerified":true,"priceSourceUrl":"https://tinker-docs.thinkingmachines.ai/tinker/models/","scoreSource":"Artificial Analysis","note":"First open-weight model from Mira Murati's Thinking Machines, released July 15, 2026. Apache 2.0, 975B total/41B active MoE, the most permissive license among frontier-scale open-weight entrants. Artificial Analysis Intelligence Index 42, re-read 2026-08-07 (up from 41); the leaderboard publishes a single configuration for it, with no effort-variant label. No independent coding composite published yet, so 'coding' is null. Thinking Machines itself frames it as not the strongest model available, open or closed, but a customizable fine-tuning base for its Tinker platform. Priced rate is a limited-time launch rate; Tinker's own docs show prefill/sample prices rising roughly 50% on 2026-07-17."},{"provider":"xAI","model":"Grok 4","origin":"United States","available":true,"inputPerMtok":3,"outputPerMtok":15,"blended":6,"coding":null,"intelligence":33.3,"codingValue":null,"intelligenceValue":5.6,"priceSource":"xAI (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://docs.x.ai/docs/models","scoreSource":"","note":"xAI reasoning model, standard sub-128K rate (price steps up above 128K). Intelligence composite only."},{"provider":"xAI","model":"Grok 4.3","origin":"United States","available":true,"inputPerMtok":1.25,"outputPerMtok":2.5,"blended":1.6,"coding":35.2,"intelligence":24.8,"codingValue":22.5,"intelligenceValue":15.9,"priceSource":"Amazon Bedrock (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://docs.x.ai/docs/models","scoreSource":"","note":"Rate shown is the Amazon Bedrock endpoint price reported by the source dataset, not xAI direct; treat as indicative pending verification."},{"provider":"Zhipu","model":"GLM-5.2","origin":"China","available":true,"inputPerMtok":1.4,"outputPerMtok":4.4,"blended":2.2,"coding":46.5,"intelligence":34.8,"codingValue":21.6,"intelligenceValue":16.2,"priceSource":"Z.ai via OpenRouter (verified)","priceVerified":true,"priceSourceUrl":"https://openrouter.ai/z-ai/glm-5.2","scoreSource":"Price Per Token, re-read 2026-09-04","note":"Open-weights flagship, superseded by GLM-5.3 and GLM-5.3-Flash. Its composite was revised sharply downward between readings: coding 68.8 to 46.5 and intelligence 51.1 to 34.8 between 2026-07-13 and 2026-09-04, on the same source. It no longer ties Gemini 3.1 Pro on coding. Still the default model in the published Hermes Agent catalog."},{"provider":"Alibaba","model":"Qwen3.8 Max","origin":"China","available":true,"inputPerMtok":2,"outputPerMtok":6,"blended":3,"coding":null,"intelligence":58,"codingValue":null,"intelligenceValue":19.3,"priceSource":"Alibaba_Qwen announcement, measured by Artificial Analysis","priceVerified":false,"priceSourceUrl":"https://artificialanalysis.ai/models/qwen3-8-max","scoreSource":"Artificial Analysis","note":"Alibaba's 2.4T-parameter flagship, announced 2026-08-03. Artificial Analysis Intelligence Index 58, read 2026-08-07, with no effort-variant label published, so this is the only configuration on the board. It is the first independent measurement of any Qwen3.8 model: an 11-point gain over Qwen3.7 Max, but still below Kimi K3 at 60 on the same-day basis. priceVerified is false because Alibaba Cloud's own Model Studio pricing page still does not list the model; the $2/$6 rate comes from the Qwen team's X announcement and is corroborated by Artificial Analysis measuring that rate on a live Alibaba Cloud endpoint. No coding composite exists upstream, so coding is null rather than backfilled from Alibaba's own capability claims. The economics are the story: the blended price fell 26% against Qwen3.7 Max ($1.18 from $1.60) while the cost to run the full Intelligence Index rose 64% ($1,741.41 from $1,063.86), and measured output speed fell from 201.9 to 67.6 tokens per second."},{"provider":"Alibaba","model":"Qwen3.7 Max","origin":"China","available":true,"inputPerMtok":2.5,"outputPerMtok":7.5,"blended":3.8,"coding":66,"intelligence":46,"codingValue":17.6,"intelligenceValue":12.3,"priceSource":"Alibaba Model Studio (verified)","priceVerified":true,"priceSourceUrl":"https://www.alibabacloud.com/help/en/model-studio/model-pricing","scoreSource":"","note":"Canonical International (Singapore) rate. A 50% promo to $1.25/$3.75 runs to July 23, 2026; the canonical rate is used here."},{"provider":"Alibaba","model":"Qwen3.7 Plus","origin":"China","available":true,"inputPerMtok":0.32,"outputPerMtok":1.28,"blended":0.6,"coding":55.9,"intelligence":39,"codingValue":99.8,"intelligenceValue":69.6,"priceSource":"Together (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://www.alibabacloud.com/help/en/model-studio/model-pricing","scoreSource":"","note":"The cost-effective Qwen3.7 tier. Rate shown is a Together endpoint price from the source dataset, not Alibaba direct; very cheap, so it ranks high on value."},{"provider":"Moonshot","model":"Kimi K2.7 Code","origin":"China","available":true,"inputPerMtok":0.95,"outputPerMtok":4,"blended":1.7,"coding":60.8,"intelligence":41.9,"codingValue":35.5,"intelligenceValue":24.5,"priceSource":"Moonshot (verified)","priceVerified":true,"priceSourceUrl":"https://platform.moonshot.ai/docs/pricing","scoreSource":"","note":"Moonshot direct rate (not a reseller rate). Coding-agent specialist."},{"provider":"Moonshot","model":"Kimi K3","origin":"China","available":true,"inputPerMtok":3,"outputPerMtok":15,"blended":6,"coding":null,"intelligence":60,"codingValue":null,"intelligenceValue":10,"priceSource":"Moonshot (verified)","priceVerified":true,"priceSourceUrl":"https://platform.kimi.ai/docs/pricing/chat-k3","scoreSource":"Artificial Analysis","note":"Moonshot's flagship, launched 2026-07-16 at frontier-tier API pricing ($3/$15 per Mtok), well above the K2 line. Artificial Analysis Intelligence Index 60 at the Max Effort variant, re-read 2026-08-07: a 3-point rise from the 57 this board carried from 2026-07-17, the largest single move in the index-wide drift and the reason the stale figure would have ranked it below Qwen3.8 Max. The old (#4 of 189) rank claim is dropped because Artificial Analysis counts effort variants as separate entries, so overall ranks move without any model changing. It remains the highest-scoring open-weight model on the index. No independent coding composite published yet, so coding is null rather than backfilled from vendor benchmarks. Artificial Analysis lists it as open weights as of 2026-08-07, so the 2026-07-27 weight release appears to have landed; the blended price it measures is $2.31 per Mtok and the full Intelligence Index costs $2,425.11 to run."},{"provider":"MiniMax","model":"MiniMax M3","origin":"China","available":true,"inputPerMtok":0.3,"outputPerMtok":1.2,"blended":0.5,"coding":58.6,"intelligence":44.4,"codingValue":111.6,"intelligenceValue":84.6,"priceSource":"MiniMax (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://www.minimax.io/platform","scoreSource":"","note":"Author-direct rate per the source dataset. A strong coding composite at a very low price makes it the value leader."},{"provider":"DeepSeek","model":"DeepSeek V4 Flash","origin":"China","available":true,"inputPerMtok":0.44,"outputPerMtok":1.32,"blended":0.7,"coding":null,"intelligence":52,"codingValue":null,"intelligenceValue":78.8,"priceSource":"DeepSeek (verified)","priceVerified":true,"priceSourceUrl":"https://api-docs.deepseek.com/quick_start/pricing","scoreSource":"Artificial Analysis","note":"The DeepSeek-V4-Flash-0731 checkpoint, an API public beta shipped July 31, 2026 behind the same model id and price as the April build. Scored 52 at the Reasoning / Max Effort variant by Artificial Analysis, re-read 2026-08-07 (up from 50). It now ties GPT-5.6 Luna at max effort and stays 7 points above DeepSeek's own pricier V4 Pro. On Artificial Analysis's own scale it sits one point below GLM-5.2 (53 there), though this board carries GLM-5.2's Price Per Token composite of 51.1 instead, so do not read those two rows against each other. Cheapest blended price on this board, so it leads intelligence value by a wide margin. No coding composite exists for it upstream yet. Prices on this board are DeepSeek's PEAK rates from the 2026-08-16 16:00 UTC repricing, re-verified 2026-08-21. Off-peak is exactly half. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, 7 of 24 hours; every other hour is off-peak. The flat card in force until 2026-08-16 was $0.14 input and $0.28 output, so the peak rates are 3.1x and 4.7x it. This row therefore ranks lower on value per dollar than it did before 2026-08-16, which is a real price change and not a scoring change."},{"provider":"DeepSeek","model":"DeepSeek V4 Pro","origin":"China","available":true,"inputPerMtok":1.32,"outputPerMtok":3.96,"blended":2,"coding":null,"intelligence":45,"codingValue":null,"intelligenceValue":22.7,"priceSource":"DeepSeek (verified)","priceVerified":true,"priceSourceUrl":"https://api-docs.deepseek.com/quick_start/pricing","scoreSource":"Artificial Analysis","note":"V4 Pro reasoning tier, 1.6T total parameters and 49B active. Re-read from Artificial Analysis on 2026-08-07 at 45, having been corrected on 2026-07-31 from 52 to 44. Artificial Analysis publishes several configurations for this model; 45 is the Reasoning / Max Effort variant, which is the same basis as the V4 Flash 0731 score of 52 and the figure Artificial Analysis itself uses when comparing the two. The old 52 came from Artificial Analysis's April launch reporting for a differently labelled configuration, so the previous row compared across variants rather than tracking any model regression. free-models.json carries a lower figure for the same family because it reads the default-effort configuration rather than max effort; the two datasets are deliberately on different variant bases and should not be reconciled. Now scores below the cheaper V4 Flash. Intelligence composite only. Prices on this board are DeepSeek's PEAK rates from the 2026-08-16 16:00 UTC repricing, re-verified 2026-08-21. Off-peak is exactly half. Peak hours are 01:00-04:00 and 06:00-10:00 UTC, 7 of 24 hours; every other hour is off-peak. The flat card in force until 2026-08-16 was $0.435 input and $0.87 output, so the peak rates are 3.0x and 4.6x it. This row therefore ranks lower on value per dollar than it did before 2026-08-16, which is a real price change and not a scoring change."},{"provider":"Nvidia","model":"Nemotron 3 Ultra","origin":"United States","available":true,"inputPerMtok":0.6,"outputPerMtok":3.6,"blended":1.4,"coding":49.3,"intelligence":37.8,"codingValue":36.5,"intelligenceValue":28,"priceSource":"Together (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://build.nvidia.com/nvidia","scoreSource":"","note":"Open MoE reasoning model. Rate shown is a Together endpoint price from the source dataset, not Nvidia direct."},{"provider":"Mistral","model":"Devstral 2","origin":"France","available":true,"inputPerMtok":0.9,"outputPerMtok":0.9,"blended":0.9,"coding":31.3,"intelligence":19.2,"codingValue":34.8,"intelligenceValue":21.3,"priceSource":"Fireworks (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://mistral.ai/pricing","scoreSource":"","note":"Mistral's open agentic coding model (123B), from France. Rate shown is a Fireworks endpoint price from the source dataset, not Mistral direct."},{"provider":"Meta","model":"Llama 4 Maverick","origin":"United States","available":true,"inputPerMtok":0.35,"outputPerMtok":1,"blended":0.5,"coding":16.3,"intelligence":14.3,"codingValue":31.8,"intelligenceValue":27.9,"priceSource":"Parasail (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://www.llama.com/","scoreSource":"","note":"Meta's open-weight MoE model (April 2025, non-reasoning). Scores now trail the 2026 frontier; cheap, so it still ranks mid-pack on value. Rate is a Parasail endpoint price, not a Meta direct rate."},{"provider":"OpenAI","model":"GPT-5.2","origin":"United States","available":true,"inputPerMtok":1.75,"outputPerMtok":14,"blended":4.8,"coding":null,"intelligence":26,"codingValue":null,"intelligenceValue":5.4,"priceSource":"OpenAI (via Price Per Token)","priceVerified":false,"priceSourceUrl":"https://openai.com/api/pricing/","scoreSource":"","note":"Standard GPT-5.2 tier. Intelligence composite only. A useful contrast with GPT-5.2 Pro, which costs roughly 12x more for a higher score."}],"unscored":[{"model":"GPT-6 Astra","reason":"OpenAI's new flagship, released 2026-09-03 at $10/$50 per Mtok with a $1 cache read, 2.5x the GPT-5.6 Sol rate this board already scores. Artificial Analysis puts it at Intelligence Index 61 at max effort (v4.1.1, read 2026-09-04), which ties Sol at max and trails Claude Fable 5.1 at 66, Claude Opus 5 at 63 and Muse Spark 1.3 at 62, and its AA Coding Agent Index of 67 trails Fable 5.1 in Claude Code at 70. Price Per Token, the base source this board uses, has published no composite for it, and a row scored from a different index would not be comparable with the scored rows. Note also that this board's blended price formula ((3*input + output)/4) puts Astra at $20, level with Fable 5.1 and Fable 5, which hides the thing that actually separates them: on Terminal-Bench 4.0 Astra solves a task for $17.02 against Fable 5.1's $32.69 because it burns 1.53B tokens to Fable 5.1's 2.75B. That effect is modeled on /coding-agent-cost-per-task/ instead. Promote once Price Per Token publishes."},{"model":"Claude Fable 5.1","reason":"Anthropic's newest frontier model, released September 1, 2026 at $10/$50 per Mtok, the same sticker as the Fable 5 it extends. Its one price change is the cache read, cut from $1.00 to $0.25 per Mtok, which this board's blended price formula ((3*input + output)/4) does not see at all: on blended price Fable 5.1 and Fable 5 are identical at $20, so a scored row would rank it exactly where Fable 5 already sits and hide the only thing that changed. Price Per Token, the base source this board uses, lists only anthropic-claude-fable-5 as of this check and has published no composite for 5.1. Left off pending that composite, the same treatment as Claude Opus 5 and Claude Sonnet 5; the cache-read effect is modeled on /coding-agent-cost-per-task/ instead, where cache tokens are billed separately."},{"model":"GLM-5.3-Flash","reason":"Z.ai's official reveal of the Ox Alpha stealth model, released 2026-08-26 at $0.15/$0.50 per Mtok standard rate ($0.03 cache read), a 320B-A18B mixture-of-experts model, MIT-licensed. Artificial Analysis scores it Intelligence Index 57, read 2026-08-27, level with Claude Opus 4.8 (also 57, same read date) and just behind Qwen3.8 Max (58). Price Per Token, the base source this board uses, has not published a composite for it yet (only z-ai-glm-5.3 and z-ai-glm-5.2 exist in that dataset as of this check). Left off the scored rows pending that composite, the same treatment as Claude Opus 5 and Claude Sonnet 5 below; GLM-5.2 continues to represent Zhipu on the scored board for now."},{"model":"Claude Opus 5","reason":"Anthropic's new flagship, released July 24, 2026 at $5/$25 per Mtok (the same rate as Opus 4.8, which it replaces). Artificial Analysis rates it Intelligence Index 63 at max effort, re-read 2026-08-07 (up from 61), still the highest score on the index and just ahead of Fable 5 at 62 in its fallback configuration. It is also the most expensive model to evaluate: $3,836.05 to run the full Intelligence Index. Price Per Token, the base source this board uses, has not yet published a coding/intelligence composite for it. It is left off pending that composite to keep the Anthropic rows on one source; Opus 4.8 represents the tier below for now."},{"model":"Claude Sonnet 5","reason":"Released June 30, 2026. Price Per Token has not yet published an independent coding/intelligence composite for it. Artificial Analysis scores it at Intelligence Index 55 at max effort, re-read 2026-08-07 (up from 53), and Anthropic's own launch benchmarks (SWE-bench Pro 63.2%, Terminal-Bench 2.1 80.4%) are self-reported; it is left off pending a Price Per Token composite to keep the Anthropic rows on one source."},{"model":"GPT-5.5","reason":"OpenAI's prior flagship, now superseded by the GPT-5.6 family (Sol, Terra, Luna), which is scored above. Price Per Token has not published a composite for GPT-5.5 and it is no longer OpenAI's current tier, so it is not scored here."},{"model":"Meta Muse Spark 1.1","reason":"Meta's first paid model, released July 9, 2026 at $1.25/$4.25 per Mtok. Artificial Analysis scores it at Intelligence Index 53 in the Xhigh Effort configuration, read 2026-08-07. That is 10 points above the 43 this note previously carried, which is far larger than the index-wide drift of the same period; the earlier read was recorded without a variant label, so it was most likely a lower-effort configuration rather than a change in the model. Treat the 43 as unattributable and the 53 as the Xhigh figure specifically. Artificial Analysis has since added a Meta Muse Spark 1.2 at 57 in the same configuration, which this board does not yet track. Price Per Token, the base source this board uses, has not published a composite for either. Muse Spark leads on tool-use and agentic tests but trails the frontier on coding. Meta Llama 4 Maverick represents Meta on the board for now."},{"model":"Cohere North Mini Code","reason":"Free on hosted endpoints and open-weight, so a per-token value score is undefined. It posts a 33.4 Artificial Analysis Coding Index, which was not re-read on 2026-08-07 because the Coding Index is not exposed on the public models leaderboard; its Intelligence Index reads 20 there, a reminder that the two indices are different scales and must not be swapped for one another. The real cost is self-hosted compute, not a token rate."},{"model":"Smaller and older variants","reason":"Models below roughly 10B parameters, superseded 2024-era releases (Claude 3.5, GPT-4 Turbo, o1), and narrowly tracked or unpriced entries are left off to keep the board to current, recognizable, buyable models."},{"model":"Google Gemini 3.8 Flash","reason":"Released 2026-09-02 at $0.75/$3.75 per Mtok, the same rate Gemini 3.6 Flash and 3.7 Flash carry, and introductory through 2026-12-31 before doubling on 2027-01-01. Artificial Analysis scores it 59 on the Intelligence Index at high effort (read 2026-09-03). Price Per Token, the base source this board uses, has published no composite for it, and a board score built from a different index would not be comparable with the scored rows. Promote once Price Per Token publishes."},{"model":"Meta Muse Spark 1.3","reason":"Released 2026-09-02 at $1.25/$4.25 per Mtok, unchanged from Muse Spark 1.2. Artificial Analysis scores the generally available xhigh configuration 61 on the Intelligence Index and the partner-only max configuration 62 (read 2026-09-03), against 57 for 1.2 at xhigh. Price Per Token has published no composite for any Muse Spark generation, so Meta is still represented on the board by Llama 4 Maverick. Promote once Price Per Token publishes."},{"model":"Qwen3.8-Max-0902","reason":"Qwen's in-place flagship refresh, dated 2026-09-02 on the QwenCloud changelog, at the same $2/$6 per Mtok as the August Qwen3.8-Max base. Qwen's own table claims Terminal-Bench 3.0 doubled from 11.3 to 29.0, vendor-reported and unreproduced, and the comparison predates Claude Fable 5.1. Price Per Token has published no composite for the snapshot, and a board score built from a different index would not be comparable with the scored rows; Qwen3.8 Max continues to represent Alibaba on the scored board for now. Promote once Price Per Token publishes."}]}