Free LLM API Leaderboard · 2026 Live Test
Daily probes at 06:00 UTC; deep quota re-check at 03:00 UTC. Composite score = 35% availability + 15% speed + 10% headroom + 40% model capability (large-model first).
📊 Current Rankings
| # | Provider | Latency (ms) | RPM | Models | Composite Score | |
|---|---|---|---|---|---|---|
| 1 |
nvidia-nim
|
1194 | — | 82 | 93.2 | 详情 › |
具体模型 (82) — 点击详情页查看完整说明 01-ai/yi-largeadept/fuyu-8bai21labs/jamba-1.5-large-instructaisingapore/sea-lion-7b-instructbigcode/starcoder2-15bdatabricks/dbrx-instructdeepseek-ai/deepseek-coder-6.7b-instructdeepseek-ai/deepseek-v4.1-flashgoogle/codegemma-1.1-7bgoogle/codegemma-7bgoogle/deplotgoogle/diffusiongemma-26b-a4b-itgoogle/gemma-2bgoogle/gemma-3-12b-itgoogle/gemma-3-4b-itgoogle/gemma-4-31b-itgoogle/recurrentgemma-2bibm/granite-3.0-3b-a800m-instructibm/granite-3.0-8b-instructibm/granite-34b-code-instructibm/granite-8b-code-instructmeta/codellama-70bmeta/llama-3.2-11b-vision-instructmeta/llama-3.2-90b-vision-instructmeta/llama-guard-4-12bmeta/llama2-70bmeta/muse-glimmer-30bmicrosoft/kosmos-2microsoft/phi-3-vision-128k-instructmicrosoft/phi-3.5-moe-instructmistralai/codestral-22b-instruct-v0.1mistralai/mistral-7b-instruct-v0.3mistralai/mistral-largemistralai/mistral-large-2-instructmistralai/mistral-nemotronmistralai/mixtral-8x22b-v0.1moonshotai/kimi-k2.6moonshotai/kimi-k3nv-mistralai/mistral-nemo-12b-instructnvidia/ai-synthetic-video-detectornvidia/cosmos-reason2-8bnvidia/embed-qa-4nvidia/ising-calibration-1.5-31bnvidia/llama-3.1-nemoguard-8b-content-safetynvidia/llama-3.1-nemoguard-8b-topic-controlnvidia/llama-3.1-nemotron-51b-instructnvidia/llama-3.1-nemotron-70b-instructnvidia/llama-3.1-nemotron-safety-guard-8b-v3nvidia/llama-3.1-nemotron-ultra-253b-v1nvidia/llama-3.2-nemoretriever-1b-vlm-embed-v1nvidia/llama-3.2-nv-embedqa-1b-v1nvidia/llama-nemotron-embed-vl-1b-v2nvidia/llama3-chatqa-1.5-70bnvidia/mistral-nemo-minitron-8b-8k-instructnvidia/nemotron-3-embed-1bnvidia/nemotron-3-nano-omni-30b-a3b-reasoningnvidia/nemotron-3-super-120b-a12bnvidia/nemotron-3-ultra-550b-a55bnvidia/nemotron-3.5-content-safetynvidia/nemotron-3.5-lightning-30b-a3bnvidia/nemotron-4-340b-instructnvidia/nemotron-4-340b-rewardnvidia/nemotron-nano-3-30b-a3bnvidia/nemotron-parsenvidia/nemotron-parse-2.0nvidia/neva-22bnvidia/nv-embedqa-mistral-7b-v2nvidia/nvclipnvidia/riva-translate-4b-instructnvidia/riva-translate-4b-instruct-v1.1nvidia/riva-translate-4b-instruct-v2nvidia/vilaopenai/gpt-oss-20bpoolside/laguna-xs-2.1snowflake/arctic-embed-lwriter/palmyra-creative-122bwriter/palmyra-fin-70b-32kwriter/palmyra-med-70bwriter/palmyra-med-70b-32kz-ai/glm-5.3z-ai/glm-5.3-flashzyphra/zamba2-7b-instruct | ||||||
| 2 |
openrouter
|
621 | — | 19 | 92.1 | 详情 › |
具体模型 (19) — 点击详情页查看完整说明 cohere/north-mini-code:freedots-studio/dots-3-note-preview:freegoogle/gemma-4-26b-a4b-it:freegoogle/gemma-4-31b-it:freeinclusionai/ling-3.0-flash-fin:freeinclusionai/ling-3.0-flash-sante:freeliquid/lfm-2.5-2.6b:freenex-agi/nex-n2.5-mini:freenex-agi/nex-n2.5-pro:freenvidia/nemotron-3-nano-omni-30b-a3b-reasoning:freenvidia/nemotron-3-super-120b-a12b:freenvidia/nemotron-3-ultra-550b-a55b:freenvidia/nemotron-3.5-content-safety:freenvidia/nemotron-3.5-lightning:freepoolside/laguna-s-2.1:freepoolside/laguna-xs-2.1:freethinkingmachines/inkling-small:freethinkingmachines/inkling:freez-ai/glm-5.2:free | ||||||
| 3 |
groq
|
534 | 1000 | 10 | 81.2 | 详情 › |
具体模型 (10) — 点击详情页查看完整说明 allam-2-7bcanopylabs/orpheus-arabic-saudicanopylabs/orpheus-v1-englishmeta-llama/llama-prompt-guard-2-22mmeta-llama/llama-prompt-guard-2-86mopenai/gpt-oss-120bopenai/gpt-oss-20bopenai/gpt-oss-safeguard-20bwhisper-large-v3whisper-large-v3-turbo | ||||||
| 4 |
agnes-free
|
5879 | — | 12 | 80.2 | 详情 › |
具体模型 (12) — 点击详情页查看完整说明 agnes-2.0-flashagnes-2.5-flashagnes-2.5-proagnes-2.5-pro-alphaagnes-2.5-pro-betaagnes-3.0-flashagnes-image-2.0-flashagnes-image-2.1-flashagnes-image-2.5-flashagnes-video-2.5agnes-video-2.5-flashagnes-video-v2.0 | ||||||
| 5 |
cohere-trial
|
451 | 20 | 13 | 69.5 | 详情 › |
具体模型 (13) — 点击详情页查看完整说明 c4ai-aya-expanse-32bcommand-a-03-2025command-a-plus-05-2026command-a-reasoning-08-2025command-r-08-2024command-r-plus-08-2024command-r7b-12-2024command-r7b-arabic-02-2025embed-english-light-v3.0embed-english-v3.0embed-multilingual-light-v3.0embed-multilingual-v3.0embed-v4.0 | ||||||
| 6 |
cloudflare-free
|
1534 | — | 2 | 68.7 | 详情 › |
具体模型 (2) — 点击详情页查看完整说明 @cf/meta/llama-3.3-70b-instruct-fp8-fast@cf/openai/gpt-oss-120b | ||||||
| 观察中 — 可用性 0（当前不可用，不建议生产使用） | ||||||
| 0 |
mistral
|
285 | — | 35 | 40.4 | 详情 › |
具体模型 (35) — 点击详情页查看完整说明 codestral-2508codestral-latestdevstral-2512devstral-latestdevstral-medium-latestlabs-leanstral-1-5labs-leanstral-1-5-1magistral-medium-latestmagistral-small-latestministral-14b-2512ministral-14b-latestministral-3b-2512ministral-3b-latestministral-8b-2512ministral-8b-latestmistral-code-agent-latestmistral-code-fim-latestmistral-code-latestmistral-mediummistral-medium-2505mistral-medium-2508mistral-medium-2604mistral-medium-3mistral-medium-3-5mistral-medium-3.5mistral-medium-latestmistral-small-2603mistral-small-latestmistral-vibe-cli-fastmistral-vibe-cli-latestmistral-vibe-cli-with-toolsvoxtral-mini-2602voxtral-mini-latestvoxtral-small-2507voxtral-small-latest | ||||||
| 0 |
gemini
|
564 | — | 14 | 38.4 | 详情 › |
具体模型 (14) — 点击详情页查看完整说明 gemini-3-flash-previewgemini-3.5-flashgemini-3.6-flashgemini-3.7-flashgemini-3.1-flash-litegemini-3.1-flash-lite-previewgemini-flash-latestgemini-flash-lite-latestgemma-4-26b-a4b-itgemma-4-31b-itgemini-3.1-pro-previewgemini-3.1-pro-preview-customtoolsgemini-3.5-flash-litegemini-3.8-flash | ||||||
📝 Daily Changelog
2026-08-22
【更正】Gemini 渠道模型计数统一为实测入库口径（28→10）：此前榜单页展示的 Gemini 模型数（主榜16+专项12=28）系 key 验证期候选分析口径，「模型数30」则来自上游 /v1beta/models 列表重复计数，均未实际写入渠道池。经生产库直读 + 全模型逐个冒烟复核，Gemini 渠道池实际入库 10 个对话文本模型（8 个 gemini text + 2 个 gemma），10/10 实测 HTTP 200。全站文章已同步更正：模型表改为 10 模型权威清单，专项区块移除，演示型号切换为 gemini-3.6-flash；gemini-2.5-* 对新用户返回 404 已下线并标注。
[Correction] Gemini channel model count unified to verified provisioned list (28→10): the previously displayed Gemini model count on the leaderboard (16 main + 12 specialized = 28) was a candidate-analysis snapshot from the key-validation phase, while the "30 models" figure came from duplicate counting of the upstream /v1beta/models listing; neither was ever provisioned into the channel pool. Verified via production DB direct read plus per-model smoke tests, the Gemini pool actually contains 10 conversational text models (8 gemini text + 2 gemma), all returning HTTP 200. All site articles have been updated accordingly: the model table now lists the authoritative 10-model roster, the specialized section is removed, the demo model is switched to gemini-3.6-flash, and gemini-2.5-* (404 for new users) has been delisted with an explicit note.
2026-08-22
Gemini 与 Mistral 两渠道正式入池上线：Gemini 10 个对话/文本模型实测入库并进入主榜综合评分（旗舰 gemini-3.7-flash、免费层最稳的 gemini-3.6-flash、gemma-4 开源双子等，Flash 系列最高 1M 输入 / 64K 输出；图像/TTS/向量等专项能力免费层配额为 0，未入池）；Mistral 39 个 chat 类模型全量入池（mistral-large/medium/small、magistral、devstral、ministral、codestral 等），9 个 voxtral-* 音频模型标 audio 进专项区。详见 《Google Gemini 免费层》 与 《Mistral 免费 API》。
Gemini and Mistral officially joined the channel pool: 10 Gemini chat/text models verified into the main ranking (flagship gemini-3.7-flash, free-tier-stable gemini-3.6-flash, the open-source Gemma pair, Flash series up to 1M-token input / 64K output; specialist capabilities like image/TTS/embedding carry zero free-tier quota and are not pooled); Mistral ships 39 chat-class models into the pool (mistral-large/medium/small, magistral, devstral, ministral, codestral, etc.) plus 9 voxtral-* audio models tagged audio in the specialist block. See the Google Gemini Free Tier guide and the Mistral Free API guide.
2026-08-22
Groq 渠道池 13 个模型逐一实测通过并全量上线：主榜新增 openai/gpt-oss-120b、openai/gpt-oss-20b、qwen/qwen3.6-27b、groq/compound、groq/compound-mini、allam-2-7b；专项区新增 whisper-large-v3(-turbo)、orpheus TTS×2（待 Groq 条款激活）、gpt-oss-safeguard-20b、llama-prompt-guard×2；旧 llama-3.x 已移出渠道池。详见《Groq 免费极速推理 API》。
Groq channel pool fully verified: 13 models tested live. Main ranking adds openai/gpt-oss-120b, openai/gpt-oss-20b, qwen/qwen3.6-27b, groq/compound, groq/compound-mini, allam-2-7b; specialist block adds whisper-large-v3(-turbo), Orpheus TTS ×2 (pending Groq ToS activation), gpt-oss-safeguard-20b, llama-prompt-guard ×2; legacy llama-3.x removed from the pool. See the Groq guide for details.
Every channel change is recorded here after live verification.
📈 Latency Trend (ms) — lower is better
Data source: /api/free-llm-rankings/history (JSON, consumed by the chart).