{ "_comment": "Verified free-tier / rate-limit table for programmatic LLM API access. Every entry re-verified against the provider's OWN public docs on the as_of date — no third-hand listicle numbers. 'unpublished' means the provider hides the figures behind a login; we say so instead of quoting rumors. Free tiers are the most volatile numbers in this repo: if a row drifted, open an issue.", "as_of": "2026-08-31", "review_note": "本轮复核 2026-08-31(上一版 2026-07-29)。**这是部分复核,不是全量重验**:实际回查了 10/17 条记录,依据各条自带的 sources 重新取证。结果:2 条已更正、7 条经取证确认不变、2 条取证后仍无法判定。 **未复核(7 项)**:Google Gemini API (AI Studio)、Hugging Face Inference Providers、Alibaba Cloud Model Studio (international)、Together AI、Moonshot / Kimi (platform.kimi.ai)、Chutes.ai、xAI (Grok API)——这些条目的值仍是上一版的,下一轮优先补。 GitHub Models 退役经两源独立复核:厂商文档明载 2026-07-30,且 models.github.ai 推理端点返回 HTTP 410 Gone。", "free_tiers": [ { "provider": "OpenRouter (:free models)", "free_offer": "Any model ID ending in :free costs $0 with a plain account", "limits": "20 req/min shared across all :free models; 50 req/day if lifetime credit purchases < $10, 1,000 req/day with a one-time $10+ top-up", "models": "rotating :free pool (varies; see openrouter.ai/models?q=free)", "card_required": "no (optional one-time $10 top-up unlocks 1,000 req/day)", "catch": "free routes go to third-party providers that may train on your data per their own policies; OpenRouter exposes separate free/paid routing settings — check them", "sources": [ "https://openrouter.ai/docs/api-reference/limits", "https://openrouter.ai/docs/features/privacy-and-logging" ], "confidence": "verified" }, { "provider": "Google Gemini API (AI Studio)", "free_offer": "Flash-class models free of charge on the API free tier", "limits": "per-model RPM/RPD no longer published publicly — visible only in AI Studio after login (aistudio.google.com/rate-limit); measured in RPM / input TPM / RPD, per project; RPD resets midnight Pacific; preview/experimental models throttled tighter", "models": "Gemini 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, 3 Flash Preview, 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite (+preview), Gemma 4, Gemini Embedding 2, plus several Live/TTS previews. Note 2.5 Pro does carry a free tier; the 2.0 Flash family is gone (shut down 2026-06-01).", "card_required": "no (Tier 1 upgrade requires a billing account)", "catch": "free-tier content is 'used to improve our products' (pricing-page wording); paid tier is not", "sources": [ "https://ai.google.dev/gemini-api/docs/rate-limits", "https://ai.google.dev/gemini-api/docs/pricing" ], "confidence": "verified (tier structure and data-use wording; exact numbers are login-walled by Google)" }, { "provider": "Groq (GroqCloud Free Plan)", "free_offer": "All hosted open-weight models usable on the Free Plan with per-model caps", "limits": "per-model, org-level: llama-3.3-70b 30 RPM / 1K RPD / 12K TPM / 100K TPD · gpt-oss-120b 30 / 1K / 8K / 200K · llama-3.1-8b-instant 30 / 14.4K / 6K / 500K · qwen3.6-27b (preview) 30 / 1K / 8K / 200K", "models": "production: GPT-OSS-120B/20B, Llama-3.3-70B, Llama-3.1-8B-Instant, Whisper Large V3 (+turbo); preview: Qwen3.6-27B, MiniMax-M2.7, GPT-OSS-Safeguard-20B, Orpheus TTS, Llama-Prompt-Guard-2, compound(-mini). ⚠️ Llama-4-Scout and Qwen3-32B were removed from the catalog since 2026-07-09.", "card_required": "no — free plan works without one; the billing page documents only the upgrade path, so read this as 'no card needed to start', not as an explicit vendor promise", "catch": "daily token caps burn fast on 70B+ models (100K-200K tokens/day); positioned as a trial", "sources": [ "https://console.groq.com/docs/rate-limits", "https://console.groq.com/docs/billing-faqs" ], "confidence": "verified" }, { "provider": "Cerebras Inference (Free Trial tier)", "free_offer": "Free Trial tier, flat limits across listed models", "limits": "5 RPM / 30K TPM / 1M tokens per hour / 1M tokens per day, org-level", "models": "gpt-oss-120b, gemma-4-31b (zai-glm-4.7 removed — its scheduled 2026-08-17 deprecation has passed and it no longer appears in the live Free Trial limits table)", "card_required": "yes (changed) — docs: \"If you skip adding a payment method at sign-up, Playground and API access remain inactive until you do\"; \"$5 in free credits after adding a verified payment method\", credits expire 30 days after grant. ⚠️ cerebras.ai/pricing contradicts this (\"$5 in free credits after making an account\") — marketing is the weaker claim; trust the docs.", "catch": "5 RPM is single-user interactive only; 'access to all models' marketing vs 3 models on the actual limits table; paid Developer tier starts at $10 with ~10x limits", "sources": [ "https://inference-docs.cerebras.ai/support/rate-limits", "https://www.cerebras.ai/pricing", "https://inference-docs.cerebras.ai/models/overview" ], "confidence": "verified (limits and model list)" }, { "provider": "Cloudflare Workers AI", "free_offer": "10,000 neurons/day of inference included in Workers Free and Paid plans", "limits": "10K neurons/day hard daily allocation; per-model neuron rates published (e.g. llama-3.2-1b 2,457 neurons/M input tokens; gpt-oss-120b 31,818/M input, 68,182/M output) — roughly 4M input tokens/day on a 1B model or ~314K input tokens/day on gpt-oss-120b (derived arithmetic from official rates)", "models": "Llama 3.3 70B (fp8-fast), Llama 3.1 8B, GPT-OSS-120B/20B, Mistral 7B, DeepSeek-R1-distill-qwen-32b", "card_required": "not stated (Workers Free signup is generally card-free)", "catch": "neurons meter compute, so output tokens burn 5-10x faster than input; hard daily cutoff on the Free plan", "sources": [ "https://developers.cloudflare.com/workers-ai/platform/pricing/" ], "confidence": "verified (allocation + per-model rates official; token translations are our division)" }, { "provider": "Mistral La Plateforme (Free mode)", "free_offer": "Free API mode by default, no card, for testing/prototyping", "limits": "unpublished — docs state the dimensions (req/s, tokens/min, tokens/month) but actual values are shown per-account in the Admin Console; widely-cited figures (1 RPS / 500K TPM / 1B tokens-month) are NOT on any public official page", "models": "officially unscoped in public docs; third parties report all API models rate-limited free (unverified scope)", "card_required": "no (official: 'API access is enabled by default with no credit card required')", "catch": "free-plan data is used for model training BY DEFAULT — opt out via the \"Anonymous improvement data\" toggle in Admin Console → Privacy. Scale-plan (API) and Vibe Teams/Enterprise customers are opted out by default.", "sources": [ "https://docs.mistral.ai/getting-started/quickstarts/studio/activate-and-generate-api-key", "https://docs.mistral.ai/getting-started/quickstarts/studio/activate-and-generate-api-key", "https://help.mistral.ai/en/articles/455207-can-i-opt-out-of-my-input-or-output-data-being-used-for-training" ], "confidence": "mixed (free mode / no card / default-training verified; exact numbers login-walled)" }, { "provider": "Cohere (trial keys)", "free_offer": "Free trial (evaluation) API key, no card", "limits": "1,000 API calls/month total; per-endpoint trial rates: Chat 20 RPM, Rerank 10 RPM, Embed 2,000 inputs/min", "models": "Command A family, Command R+/R/R7B, Embed, Rerank", "card_required": "no", "catch": "evaluation scale only; the same 1,000/month cap also applies to production keys on the newest Chat variants (sales contact to lift)", "sources": [ "https://docs.cohere.com/docs/rate-limits" ], "confidence": "verified" }, { "provider": "SambaNova Cloud", "free_offer": "Free tier applied automatically when no payment method is linked", "limits": "20 RPM / 20 requests per day / 200K tokens per day, uniform across free-tier models", "models": "DeepSeek-V3.1, Llama-3.3-70B, gpt-oss-120b (production); DeepSeek-V3.2, gemma-4-31B (preview)", "card_required": "no", "catch": "20 requests/day is demo-only despite the very fast inference; Developer tier (card linked) lifts to 20M tokens/day", "sources": [ "https://docs.sambanova.ai/docs/en/models/rate-limits" ], "confidence": "verified" }, { "provider": "Hugging Face Inference Providers", "free_offer": "Monthly USD micro-credits usable across 200+ provider-routed models at passthrough prices", "limits": "$0.10/month included credits on free accounts ('subject to change'); PRO $2.00/month; Team/Enterprise $2.00 per seat pooled", "models": "whatever the integrated providers serve (DeepSeek-V3, Llama, FLUX.1-dev, ...)", "card_required": "no (card only for pay-as-you-go top-up)", "catch": "$0.10 is a handful of big-model requests; credits apply only to HF-routed requests, not custom provider keys", "sources": [ "https://huggingface.co/docs/inference-providers/en/pricing" ], "confidence": "verified" }, { "provider": "Z.ai (Zhipu GLM, international)", "free_offer": "GLM Flash models priced Free for input, cached input, AND output", "limits": "unpublished — per-key limits visible after login (z.ai/manage-apikey/rate-limits); concurrency 'dynamically adjusted'; third-party figures (~1 RPS / ~1K RPD) unverified", "models": "GLM-4.7-Flash (text), GLM-4.5-Flash (text), GLM-4.6V-Flash (vision)", "card_required": "not stated", "catch": "opaque, shifting concurrency limits (single-in-flight throttling reported even on paid tiers); China-headquartered provider — weigh your data sensitivity", "sources": [ "https://docs.z.ai/guides/overview/pricing" ], "confidence": "verified (free pricing); rate limits unpublished" } ], "trial_credits_not_free_tiers": [ { "provider": "NVIDIA build.nvidia.com (NIM APIs)", "note": "the FAQ now promises only \"free credits ... for application prototyping\" with NO number; the widely-quoted 1,000/+4,000-request figures are staff-forum answers, never documented. The 90-day NVIDIA AI Enterprise evaluation license is still documented.", "sources": [ "https://docs.api.nvidia.com/nim/docs/faq", "https://forums.developer.nvidia.com/t/nim-api-credits/305703" ], "confidence": "mixed (mechanics official; credit numbers staff-forum)" }, { "provider": "Alibaba Cloud Model Studio (international)", "note": "new-user per-model free token quotas, valid \"30 to 90 days\" from activation (new activations move to 90-day validity from 2026-09-08 03:00 UTC), Singapore region only, real-time inference only; an account and its RAM users share one quota.", "sources": [ "https://www.alibabacloud.com/help/en/model-studio/new-free-quota" ], "confidence": "verified (offer + expiry; per-model amounts not publicly extractable)" } ], "discontinued_or_never_free": [ { "provider": "GitHub Models", "note": "retired 2026-07-30 — GitHub's own docs: 'As of July 30, 2026, GitHub Models has been fully retired. The playground, model catalog, inference API, and bring your own key (BYOK) are no longer available to any customer.' Verified independently: the inference endpoint models.github.ai returns HTTP 410 Gone. Note that github.com/marketplace/models still answers 200, so a reader clicking around may not realise the service is dead. GitHub points new projects at Azure AI Foundry. Previously listed here as a live free tier (15 RPM / 150 RPD low-tier, no card) — that offer no longer exists.", "sources": [ "https://docs.github.com/en/github-models/use-github-models/prototyping-with-ai-models" ], "confidence": "high — vendor doc states the retirement date and the API endpoint returns 410" }, { "provider": "Together AI", "note": "no free models and no free trial as of 2026-07 — official docs: 'Together AI does not currently offer free trials'; $5 minimum prepaid purchase required; the legacy -free endpoints (Llama-Vision-Free, DeepSeek-R1-Distill-70B-free) are gone from the models list. Ignore stale listicles.", "sources": [ "https://docs.together.ai/docs/billing-credits", "https://docs.together.ai/docs/serverless/models" ], "confidence": "verified (negative finding)" }, { "provider": "Moonshot / Kimi (platform.kimi.ai)", "note": "no free tier — official limits page: 'you need to recharge at least $1 to start using'; Tier0 ($1) = 3 RPM / 500K TPM / 1.5M TPD", "sources": [ "https://platform.kimi.ai/docs/pricing/limits" ], "confidence": "verified (negative finding)" }, { "provider": "Chutes.ai", "note": "no free tier as of 2026-07 (verified on the official pricing page): plans are Plus $10/mo (6% off PAYG), Pro $20/mo (10% off PAYG), Enterprise. The earlier \"$3/month\" figure is stale; the claim that a $5-deposit free daily allowance was discontinued remains community-reported, not documented.", "sources": [ "https://chutes.ai" ], "confidence": "reported (pricing pages via search; not directly fetched)" }, { "provider": "xAI (Grok API)", "note": "the much-cited $150/month data-sharing credits program is no longer documented on any public official page (the docs.x.ai data-sharing page 404s; billing FAQ describes prepaid credits only) — we won't print numbers we can't verify; check console.x.ai after login", "sources": [ "https://docs.x.ai/developers/faq/billing" ], "confidence": "unverified (official pages omit or contradict third-party claims)" } ] }