{"asOf":"2026-09-05","methodology":"A model is listed as a good free option only if it is callable at $0 on at least one hosted provider and clears the quality gate: an Artificial Analysis Intelligence Index of at least 20 of 100. That index (v4.1) is the one benchmark scored for every model here; it blends nine independent evaluations including coding, so a single number ranks the table consistently. Benchmark scores are scraped from the public Artificial Analysis leaderboard; free-tier terms are verified against each provider source and dated.","qualityGate":{"metric":"Artificial Analysis Intelligence Index","min":20,"note":"the index is demanding: on v4.2 the best model of any price scores 57 and the strongest free model here scores 53, so the 0-100 scale runs low. A free model scoring at least 20 is genuinely capable. The v4.2 recalibration pushed several previously listed free models below the bar, and they are no longer ranked: OpenAI gpt-oss-120b (16), NVIDIA Nemotron 3 Super (19), Google Gemma 4 26B A4B (19), NVIDIA Nemotron 3.5 Lightning (16) and Alibaba Qwen3 Next 80B A3B (8). They are still free; they are just no longer models we would tell you to build on."},"benchmarkSource":{"name":"Artificial Analysis","url":"https://artificialanalysis.ai/leaderboards/models","metric":"Artificial Analysis Intelligence Index","version":"v4.2","note":"The Artificial Analysis Intelligence Index (v4.2, current since September 2026) is a single 0-100 score blending ten independent evaluations weighted across four categories: Agents 30% (AA-Briefcase 15%, GDPval-AA v2 10%, t3-Banking 5%), Coding 20% (Terminal-Bench v2.1 10%, SciCode 10%), Scientific Reasoning 20% (HLE 10%, CritPt 10%) and General 30% (AA-Omniscience Accuracy 10% and Non-Hallucination 5%, GDP.pdf 10%, AA-LCR v1.1 5%). It is the one benchmark that scores every model on this page. Read from the public leaderboard on the date below; this site does not run the evaluations. Two cautions. v4.2 is NOT comparable with the v4.1 scores this page carried until 2026-09-05: v4.2 added AA-Briefcase and GDP.pdf, removed GPQA Diamond, regraded SciCode and rebalanced every weight, which moved most scores down by several points. And where a lab publishes several reasoning-effort settings, the figure here is the highest one Artificial Analysis publishes, with the setting named in the model note.","scoresVerified":"2026-09-05"},"models":[{"model":"Muse Spark 1.3","maker":"Meta","aaIndex":53,"contextLength":1048576,"openWeight":false,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"The highest-scoring free model on this page by a wide margin, at max reasoning effort. Free through the OpenCode Zen Contributor tier, and that tier is the price: OpenCode Zen states you grant permission to use your prompts and completions to train future Meta models. Closed weights. The paid Muse Spark 1.3 on the same gateway lists at $1.25 and $4.25 per million tokens."},{"model":"Muse Spark 1.2","maker":"Meta","aaIndex":47,"contextLength":1048576,"openWeight":false,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"The previous Muse Spark, at xhigh effort, still free on the same OpenCode Zen Contributor tier and on the same training-data terms. Worth knowing about because it is only six points behind 1.3 and both are free."},{"model":"GLM-5.2","maker":"Z AI","aaIndex":43,"contextLength":256000,"openWeight":true,"hosts":[{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"}],"note":"Max reasoning effort. Open weights, so it can be self-hosted rather than only called. Free as z-ai/glm-5.2:free on OpenRouter, which caps context at 256K on that route."},{"model":"DeepSeek V4 Flash 0731","maker":"DeepSeek","aaIndex":41,"contextLength":200000,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"Max reasoning effort. Free on OpenCode Zen, where the free endpoint caps context at 200K. Note that paid DeepSeek moved to peak and off-peak billing on 2026-08-16, so the paid rate you compare against depends on the hour."},{"model":"Gemini 3.5 Flash","maker":"Google","aaIndex":38,"contextLength":1048576,"openWeight":false,"hosts":[{"key":"google-ai-studio","name":"Google AI Studio","accessUrl":"https://aistudio.google.com/"}],"note":"Medium reasoning effort. The only frontier-lab model with a first-party free tier on this page. Free-tier availability was last confirmed 2026-06-28: Google has since moved per-model free rate limits off its public docs page and into the AI Studio dashboard, so the exact limits can no longer be cited from a public source."},{"model":"MiniMax M3","maker":"MiniMax","aaIndex":36,"contextLength":1048576,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"},{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"}],"note":"Open weights, and free on two independent gateways, which makes it the most resilient pick here: if one throttles, the other is unaffected. OpenRouter serves the full 1M context; the OpenCode Zen free endpoint caps at 200K."},{"model":"Hy3","maker":"Tencent","aaIndex":33,"contextLength":190000,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"Tencent Hunyuan 3, open weights, free on OpenCode Zen. The free endpoint caps context at 190K against the model spec of 256K. OpenCode Zen also carries a separate free Hy3 preview endpoint."},{"model":"Inkling","maker":"Thinking Machines","aaIndex":32,"contextLength":1048576,"openWeight":true,"hosts":[{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"}],"note":"Thinking Machines first released model, open weights, free on OpenRouter. It launched paid at $1.87 and $4.68 per million tokens, so the free route is a genuine saving rather than a stripped tier."},{"model":"Inkling Small","maker":"Thinking Machines","aaIndex":32,"contextLength":1048576,"openWeight":true,"hosts":[{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"}],"note":"The smaller Inkling scores level with the full model on the v4.2 index while being cheaper to self-host. Open weights, free on OpenRouter."},{"model":"Nemotron 3 Ultra","maker":"NVIDIA","aaIndex":30,"contextLength":1000000,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"},{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"},{"key":"nvidia-nim","name":"NVIDIA NIM (build.nvidia.com)","accessUrl":"https://build.nvidia.com/"}],"note":"NVIDIA flagship open-weight model, and the most widely available free model here: three independent hosts, including NVIDIA own endpoints. NVIDIA logs free usage for security and to improve its products."},{"model":"MiMo V2.5","maker":"Xiaomi","aaIndex":30,"contextLength":200000,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"Xiaomi open-weight workhorse, free on OpenCode Zen, where collected data may be used to improve the model during the free period."},{"model":"MiMo V2 Omni","maker":"Xiaomi","aaIndex":28,"contextLength":262144,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"The multimodal MiMo, open weights, free on OpenCode Zen. Scores within two points of MiMo V2.5 on a text-first index, so treat the multimodal capability as the reason to pick it rather than the text quality."},{"model":"Ling 3.0 Flash","maker":"InclusionAI","aaIndex":27,"contextLength":262144,"openWeight":false,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"InclusionAI workhorse, free on OpenCode Zen. Closed weights despite the open-sounding lineage. OpenCode Zen and OpenRouter also carry finance-tuned and health-tuned Ling 3.0 Flash variants free, which Artificial Analysis does not score separately."},{"model":"LongCat 2.0","maker":"LongCat","aaIndex":26,"contextLength":1000000,"openWeight":false,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"A 1M-context model free on OpenCode Zen. Closed weights. The long context is the draw; the index score is mid-table."},{"model":"MiMo V2 Flash","maker":"Xiaomi","aaIndex":26,"contextLength":262144,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"The fast, small MiMo. Open weights, free on OpenCode Zen. Scores at the February 2026 revision."},{"model":"Ring 2.6 1T","maker":"InclusionAI","aaIndex":24,"contextLength":262000,"openWeight":true,"hosts":[{"key":"opencode-zen","name":"OpenCode Zen","accessUrl":"https://opencode.ai/docs/zen/"}],"note":"A trillion-parameter open-weight model offered free on OpenCode Zen. Large on paper, mid-table on the index, which is a useful reminder that parameter count is not capability."},{"model":"Gemma 4 31B","maker":"Google","aaIndex":22,"contextLength":262144,"openWeight":true,"hosts":[{"key":"openrouter","name":"OpenRouter","accessUrl":"https://openrouter.ai/models?max_price=0"}],"note":"Google open-weight family, free on OpenRouter and small enough to run locally. The lowest-scoring model that still clears the quality gate."}],"hosts":[{"key":"google-ai-studio","name":"Google AI Studio","freeSummary":"Gemini Flash and Flash-Lite free (Pro tiers left the free tier on 2026-04-01); roughly 5-15 req/min and 20-1,500 req/day depending on model","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":true,"speedNote":"","accessUrl":"https://aistudio.google.com/","sourceUrl":"https://ai.google.dev/gemini-api/docs/rate-limits","verifiedDate":"2026-06-28","note":"Free-tier inputs may be used to improve Google products (outside the UK, CH, EEA, and EU). Gemini 2.5/3.x Pro are no longer free as of 2026-04-01. Not re-verified in the 2026-09-05 pass: Google has moved per-model free rate limits off this public docs page and into the AI Studio dashboard, so the limits above are the last publicly citable figures rather than current ones."},{"key":"openrouter","name":"OpenRouter","freeSummary":"22 models tagged :free at 20 req/min and 50 req/day (1,000/day after a one-time $10 credit purchase)","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":false,"speedNote":"","accessUrl":"https://openrouter.ai/models?max_price=0","sourceUrl":"https://openrouter.ai/docs/api-reference/limits","verifiedDate":"2026-09-05","note":"A single API gateway to free models from many labs. The current free pool, read from the OpenRouter API rather than the website (the models page is client-rendered and under-reports), includes GLM-5.2, both Inkling models, MiniMax M3, NVIDIA Nemotron 3 Ultra and Super, Google Gemma 4, poolside Laguna and Cohere North Mini Code. It is community-funded and can be throttled. Alibaba Qwen3 Next and OpenAI gpt-oss are no longer in it."},{"key":"opencode-zen","name":"OpenCode Zen","freeSummary":"31 models free through the OpenCode CLI and Desktop, including Muse Spark 1.3 and 1.2, MiMo V2.5, Nemotron 3 Ultra, MiniMax M3, DeepSeek V4 Flash and the Big Pickle stealth model","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":true,"speedNote":"","accessUrl":"https://opencode.ai/docs/zen/","sourceUrl":"https://models.dev/api.json","verifiedDate":"2026-09-05","note":"A curated coding-model gateway, and by a distance the largest free pool on this page. The free list is read from models.dev, the registry OpenCode itself ships with, because the published Zen docs page lists only a subset of it. The free tier is paid for with data: OpenCode Zen states that during the free period collected data may be used to improve the model, and the Muse Spark Contributor tiers go further, granting permission to use your prompts and completions to train future Meta models. NVIDIA endpoints log usage for security and product improvement. No request caps are published."},{"key":"nvidia-nim","name":"NVIDIA NIM (build.nvidia.com)","freeSummary":"Open models free at about 40 req/min after phone verification","requiresSignup":true,"phoneVerify":true,"dataUsedForTraining":false,"speedNote":"","accessUrl":"https://build.nvidia.com/","sourceUrl":"https://build.nvidia.com/","verifiedDate":"2026-06-28","note":"NVIDIA-hosted endpoints for open models including its own Nemotron line. Free for development after phone verification."},{"key":"groq","name":"Groq","freeSummary":"Open models free at roughly 1,000 req/day on larger models, up to 14,400 on small ones; 12K tokens/min","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":false,"speedNote":"Fastest-class inference on custom LPU hardware","accessUrl":"https://console.groq.com/","sourceUrl":"https://console.groq.com/docs/rate-limits","verifiedDate":"2026-06-28","note":"Built for speed. Free-tier request limits were tightened in 2026."},{"key":"cerebras","name":"Cerebras","freeSummary":"Open models free at 30 req/min, 14,400 req/day, 60K tokens/min","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":false,"speedNote":"Among the fastest output speeds available","accessUrl":"https://cloud.cerebras.ai/","sourceUrl":"https://inference-docs.cerebras.ai/support/pricing","verifiedDate":"2026-06-28","note":"Wafer-scale hardware delivers very high tokens-per-second. Free tier serves gpt-oss-120b and Llama models."},{"key":"cloudflare","name":"Cloudflare Workers AI","freeSummary":"10,000 neurons/day free across the model catalog (a usage credit, not a request cap)","requiresSignup":true,"phoneVerify":false,"dataUsedForTraining":false,"speedNote":"","accessUrl":"https://developers.cloudflare.com/workers-ai/","sourceUrl":"https://developers.cloudflare.com/workers-ai/platform/pricing/","verifiedDate":"2026-06-28","note":"Free daily neuron allocation runs open models (Llama, Qwen, gpt-oss) at the edge. Heavier models burn neurons faster."}],"unscored":[{"model":"Big Pickle","host":"OpenCode Zen","reason":"A stealth coding model OpenCode is running free while it collects feedback. No lab, no published weights and no benchmark score, so it cannot be ranked."},{"model":"Ox Alpha (Unlimited)","host":"OpenCode Zen","reason":"A stealth model offered free with no stated request cap. Artificial Analysis does not score it."},{"model":"Grok Code Fast 1","host":"OpenCode Zen","reason":"SpaceXAI coding model, free on OpenCode Zen. Artificial Analysis scores the Grok 4.x line but not this coding-specific model."},{"model":"Laguna S 2.1 and Laguna XS 2.1","host":"OpenCode Zen / OpenRouter","reason":"poolside open-weight coding models, free on both gateways, but Artificial Analysis does not score either."},{"model":"Ling 3.0 Flash Fin and Ling 3.0 Flash Sante","host":"OpenCode Zen / OpenRouter","reason":"Finance-tuned and health-tuned Ling 3.0 Flash variants. Artificial Analysis scores only the base Ling 3.0 Flash, so the tuned variants cannot be ranked separately."},{"model":"MiniMax M2.1, M2.5 and M2.7","host":"OpenCode Zen / OpenRouter","reason":"Earlier MiniMax releases still offered free. Artificial Analysis v4.2 scores M3 but not these."},{"model":"GLM-4.7 and GLM-5","host":"OpenCode Zen","reason":"Earlier Z AI models free on OpenCode Zen. v4.2 scores GLM-5.2 and GLM-5.3, but not these two."},{"model":"Kimi K2.5","host":"OpenCode Zen","reason":"A Moonshot open-weight model free on OpenCode Zen. v4.2 scores K3 and K2.7 Code, not K2.5."},{"model":"North Mini Code","host":"OpenCode Zen / OpenRouter","reason":"A Cohere coding model free on both gateways. Artificial Analysis scores it at 13 on v4.2, below this page quality gate, so it is listed here rather than ranked."}]}