Discover free LLM API models across providers with real speed and latency benchmarks.Free models 2773Providers 157Free offerings 8806
| Model | Providers | Speed | Latency | Tests |
|---|---|---|---|---|
IBMFree IBM Granite 4.0 H Micro is a compact language model in the Granite series, optimized for quick responses and high throughput. | N/A | N/A | 0 | |
微软Free Microsoft Phi 4 Mini Instruct is a compact instruction-tuned variant in the Phi series, optimized for quick responses and high throughput. | +5 moreShow fewer | N/A | N/A | 0 |
OpenAIFree |
LMSpeed tracks 2773 LLM models available for free across 157 API providers. Free tiers vary by provider — some offer limited daily requests, others provide free credits for new users. All speed data is from real API tests.
Find a free model, compare its providers, then open the provider or model detail page before you start testing.
Use search, model-family filters, and capability tags to narrow the directory to the free models you actually want to try.
Check provider count, benchmark volume, throughput, and first-token latency so you are not choosing from price alone.
Review the provider page for availability, health checks, pricing notes, and any free-tier limits before building on it.
Common questions about free model tiers and how to use this directory.
| N/A |
| N/A |
| 0 |
CC逆向ReasoningToolsFilesAnthropic Claude Haiku 4.5 delivers fast, low-cost responses while retaining solid instruction following for chat, classification, and lightweight coding. | 兔子API我不是AI神Moyanjdc API +44 moreShow fewerHigobs APIKternaDMXAPIPrivnode丰思理 AIVenlacyCan API枫叶17NAS API天宫造物丸美小沐写作丸美小沐MagicAI艾可APIQWQ Chat APIAIO通用智能服务平台Fangyuan APILiunew APIAIsa42公益站DeepRouterKoru APINeb 公益站ZetaTechs APIOptAISwifllyLLMTommyLam API毫秒APIMentoe APIDibin84 API HubQQ CodeNuizi APISeamee APIRinkoAICHB API南北红豆不知道叫啥APDSMMapleLeaf APIAI APIWAADRIMyNav AI10dian-API碳硅生命体 | 88.31 t/s | 2.24 s | 10 |
Qwen3-VL-8B-Instruct is a multimodal vision-language model from the Qwen3-VL series, built for high-fidelity understanding and reasoning across text, images, and video. It features improved multimodal fusion with Interleaved-MRoPE for long-horizon... | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI GPT-5 Image is an image generation model that creates high-quality visuals from text prompts, supporting creative design, marketing assets, and multimodal applications. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI O3 Deep Research is a reasoning model in the O series, designed for complex reasoning, problem-solving, and analytical tasks. | N/A | N/A | 0 |
ReasoningToolsFiles200KOpenAI O4 Mini Deep Research is a specialized reasoning model optimized for deep research and analytical tasks. | N/A | N/A | 0 |
FilesVision32.8KReasoningGoogle Gemini 2.5 Flash Image is an image generation model, capable of producing images from text prompts. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI GPT-5 Pro is a high-capability language model in the GPT-5 series, offering enhanced reasoning, code generation, and multimodal capabilities. | N/A | N/A | 0 |
ReasoningTools200KOpen WeightsZhipu AI GLM-4.6 builds on GLM-4.5 with a 200K context window, stronger real-world coding, advanced reasoning with tool use, and improved agentic performance for complex multi-step tasks. | N/A | N/A | 0 |
A\claude-codeSonnetReasoningAnthropic Claude Sonnet 4.5 balances strong reasoning, coding, and agent capabilities with lower latency and cost than Opus, making it a practical default for production assistants. | 45.07 t/s | 2.95 s | 15 |
ReasoningToolsFilesVisionGoogle Gemini 2.5 Flash Lite is a lightweight and cost-efficient language model in the Gemini series, optimized for fast responses at reduced cost. | N/A | N/A | 0 |
256KTools262.1KChatAlibaba Qwen3 Max is the largest Qwen3-tier model, targeting frontier reasoning, long-context understanding, coding, and agent tasks at maximum capability. | N/A | N/A | 0 |
ToolsOpen Weights1M256KAlibaba Qwen3 Coder Plus is a code-specialized variant in the Qwen series, optimized for code generation, debugging, and software development tasks. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI GPT-5 Codex is a code-specialized variant in the GPT-5 series, optimized for code generation, debugging, and software development tasks. | N/A | N/A | 0 |
ReasoningTools128K164KDeepSeek V3.1 Terminus is a Terminus-branded DeepSeek V3.1 variant tuned for stable long-context reasoning and code generation. | N/A | N/A | 0 |
Tools128K256K262KMoonshot Kimi K2 is a trillion-parameter MoE model from Moonshot AI, built for long-context chat, agentic tool use, and high-quality Chinese-English bilingual generation. | N/A | N/A | 0 |
DeepSeek Chat V3.1 is an instruction-tuned variant in the DeepSeek series, optimized for following instructions and conversational tasks. | N/A | N/A | 0 |
Mistral Medium 3.1 is a language model in the Mistral series, offering general-purpose reasoning, code generation, and multimodal capabilities. | N/A | N/A | 0 |
ToolsFilesVision66KZhipu AI GLM-4.5V is a multimodal vision-language model in the GLM series, supporting both text and image understanding. | N/A | N/A | 0 |
ReasoningFilesVision128KOpenAI GPT-5 is OpenAI frontier general-purpose model with improved reasoning depth, coding reliability, and multimodal understanding for production assistants and agent workflows. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI GPT-5 Mini is a compact language model in the GPT-5 series, optimized for low-latency responses and efficient inference. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI GPT-5 Nano is a compact language model in the GPT-5 series, optimized for low-latency responses and efficient inference. | N/A | N/A | 0 |
ReasoningToolsOpen Weights131.1KGPT-OSS is an open-weight language model family designed for self-hosted inference, research, and cost-efficient alternatives to proprietary GPT-class models. | Dext APIApiToken Online兔子API +38 moreShow fewerZero API初叶🍂Furry API猫羽雫API猫羽霖APIMoyanjdc APIFuture Hub霸气公益平台Elysiver APIDMXAPIAI Claw API小辣椒LLM APIMitchll-API艾可APIXiao WanAIsaDeepRouterKoru API随时跑路公益站Koyeb AI GatewaySWT-APIZetaTechs APITommyLam API毫秒APISeamee APIRinkoAICHB APIfffaa AIwuer的api站APDSMMineWuer APIKFC API星见雅 APIMapleLeaf APIAI APIWAADRIC85 API温云 | 443.53 t/s | 0.98 s | 20 |
ReasoningToolsFilesVisionAnthropic Claude Opus 4.1 is a high-capability Claude release focused on demanding reasoning, synthesis, and agent orchestration tasks. | N/A | N/A | 0 |
ReasoningToolsOpen Weights131.1KZhipu AI GLM-4.5 is a 355B-parameter MoE agent foundation model that unifies reasoning, coding, and tool use with hybrid thinking modes and a 128K context window. | 15.30 t/s | 5.31 s | 10 |
ReasoningToolsOpen Weights131.1KZhipu AI GLM-4.5 Air is a lightweight GLM-4.5 variant optimized for low-latency Chinese and English dialogue, retrieval-augmented apps, and edge deployments. | N/A | N/A | 0 |
Alibaba Qwen3 Coder is a code-specialized variant in the Qwen series, optimized for code generation, debugging, and software development tasks. | N/A | N/A | 0 |
ReasoningTools40KOpen WeightsAlibaba Qwen3 is the Qwen family's flagship LLM series with dense and MoE variants, seamless thinking/non-thinking modes, and leading open-source performance in math, code, and agent tasks. | N/A | N/A | 0 |
ChatReasoningTools131KMiniMax M1 is a foundation model in the MiniMax lineup, built for general chat, multilingual generation, and scalable API deployment across text-centric applications. | N/A | N/A | 0 |
FilesVision1MReasoningGoogle Gemini 2.5 Flash is a fast multimodal model balancing speed and intelligence for chat, tool use, and large-context workloads at lower cost than Pro tiers. | N/A | N/A | 0 |
ReasoningFilesVision1MGoogle Gemini 2.5 Pro is Google advanced multimodal model with a 1M-token context window, strong STEM reasoning, and native support for images, audio, and video understanding. | 91.91 t/s | 16.35 s | 5 |
ReasoningToolsFilesVisionOpenAI O3 Pro is a reasoning model in the O series, designed for complex reasoning, problem-solving, and analytical tasks. | N/A | N/A | 0 |
128KReasoningOpen Weights163.8KDeepSeek R1 is a reasoning-focused language model in the DeepSeek series, designed for complex reasoning, problem-solving, and analytical tasks. | 25.85 t/s | 5.73 s | 10 |
ToolsFilesVision200KAnthropic Claude Opus 4 is the most capable Claude 4 model for deep reasoning, research, complex coding, and high-stakes agent workflows requiring maximum quality. | N/A | N/A | 0 |
ReasoningToolsFilesVisionAnthropic Claude Sonnet 4 balances speed and intelligence for coding, analysis, and enterprise automation, with strong instruction following and long-context performance. | N/A | N/A | 0 |
Mistral Medium 3 is a high-performance enterprise-grade language model designed to deliver frontier-level capabilities at significantly reduced operational cost. It balances state-of-the-art reasoning and multimodal performance with 8× lower cost... | N/A | N/A | 0 |
ReasoningTools200KFilesOpenAI O4 Mini is a reasoning model in the O series, designed for complex reasoning, problem-solving, and analytical tasks. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI O3 is a reasoning model in the O series, designed for complex reasoning, problem-solving, and analytical tasks. | N/A | N/A | 0 |
ToolsFilesVision1MOpenAI GPT-4.1 delivers upgraded coding, instruction following, and long-context performance as a successor within the GPT-4 family. | 80.83 t/s | 1.82 s | 120 |
ToolsFilesVision1MOpenAI GPT-4.1 Mini is a compact language model in the GPT-4 series, optimized for low-latency responses and efficient inference. | N/A | N/A | 0 |
ToolsFilesVision1MOpenAI GPT-4.1 Nano is a compact language model in the GPT-4 series, optimized for low-latency responses and efficient inference. | 126.63 t/s | 2.01 s | 75 |
ReasoningToolsOpen WeightsVisionMeta Llama 4 Scout is a compact open-weight model in the Llama 4 family, designed for efficient inference, on-device deployment, and low-latency agent workloads. | N/A | N/A | 0 |
ReasoningToolsFilesVisionOpenAI O1 Pro is a premium reasoning model in the O series, offering enhanced reasoning capabilities for complex tasks. | N/A | N/A | 0 |
Command A is an open-weights 111B parameter model with a 256k context window focused on delivering great performance across agentic, multilingual, and coding use cases. Compared to other leading proprietary... | N/A | N/A | 0 |
FilesVision128KOpenAI GPT-4o Mini Search is a search-augmented language model in the GPT-4 series, integrating web retrieval to provide up-to-date answers. | N/A | N/A | 0 |
FilesVision128KOpenAI GPT-4o Search is a search-augmented language model in the GPT-4 series, integrating web retrieval to provide up-to-date answers. | N/A | N/A | 0 |
Reka Flash 3 is a general-purpose, instruction-tuned large language model with 21 billion parameters, developed by Reka. It excels at general chat, coding tasks, instruction-following, and function calling. Featuring a... | N/A | N/A | 0 |
Note: Sonar Pro pricing includes Perplexity search pricing. See [details here](https://docs.perplexity.ai/guides/pricing#detailed-pricing-breakdown-for-sonar-reasoning-pro-and-sonar-pro) Sonar Reasoning Pro is a premier reasoning model powered by DeepSeek R1 with Chain of Thought (CoT). Designed for... | N/A | N/A | 0 |
Note: Sonar Pro pricing includes Perplexity search pricing. See [details here](https://docs.perplexity.ai/guides/pricing#detailed-pricing-breakdown-for-sonar-reasoning-pro-and-sonar-pro) For enterprises seeking more advanced capabilities, the Sonar Pro API can handle in-depth, multi-step queries with added extensibility, like... | N/A | N/A | 0 |
Mistral Saba is a 24B-parameter language model specifically designed for the Middle East and South Asia, delivering accurate and contextually relevant responses while maintaining efficient performance. Trained on curated regional... | N/A | N/A | 0 |
ReasoningTools200KFilesOpenAI O3 Mini is a compact reasoning model in the O series, designed for efficient reasoning and problem-solving tasks. | 55.62 t/s | 10.28 s | 21 |
Aion RP Llama 3.1 is a roleplay-tuned variant in the Aion series, optimized for character-driven dialogue and creative writing. | N/A | N/A | 0 |
ReasoningTools131KReasoningDeepSeek R1 Distill Qwen is a reasoning model in the DeepSeek series, distilling R1 capabilities into the Qwen architecture for complex reasoning and analytical tasks. | N/A | N/A | 0 |
Sonar is lightweight, affordable, fast, and simple to use — now featuring citations and the ability to customize sources. It is designed for companies seeking to integrate lightweight question-and-answer features... | N/A | N/A | 0 |
DeepSeek R1 Distill Llama 70B is a distilled large language model based on [Llama-3.3-70B-Instruct](/meta-llama/llama-3.3-70b-instruct), using outputs from [DeepSeek R1](/deepseek/deepseek-r1). The model combines advanced distillation techniques to achieve high performance across... | N/A | N/A | 0 |
MiniMax-01 is MiniMax early-generation foundation model combining text and multimodal capabilities, serving as the base for the MiniMax model lineup. | N/A | N/A | 0 |
ChatMicrosoft Phi-4 is a compact open-weight language model in the Phi series, optimized for reasoning, math, code generation, and efficient on-device or API deployment. | N/A | N/A | 0 |
Reasoning128KToolsFilesOpenAI O1 is a reasoning model in the O series, designed for complex reasoning, problem-solving, and analytical tasks. | N/A | N/A | 0 |
A model–provider pair whose current input and output prices are both $0 per token. Some providers offer permanent free tiers, others only give one-time credits to new accounts. We mark an offering as free only while its public pricing is zero, and re-check pricing pages regularly.
Most have rate limits — requests per minute, daily caps, or context-length limits — and many require account signup with a verified phone or payment method. Some are time-limited promotions. Always read the provider's terms and quotas before depending on a free endpoint.
Some entries are community-run relays ("公益站") that bundle paid upstream keys and redistribute access for free at the operator's expense. They often advertise larger quotas and a broader model list than official free tiers, but reliability is much lower: operators can pull the plug or disappear overnight, pricing and quotas can change without notice, and many sites are invite-only — requiring a GitHub invite, a forum referral, or a closed community to register. Some keep signups disabled indefinitely. Treat them as best-effort backup channels; keep anything important on official paid endpoints.
Speed varies by model and provider. Sort the table by Most tested for the most reliable benchmarks, or pick a model family from the chips above to drill in. Each row shows median tokens per second and first-token latency from real API tests.
We send identical prompts to each provider through a five-round stress test, count output tokens with tiktoken, and measure both throughput (tokens per second) and time to first token. Numbers are aggregated as medians to resist outliers and refresh on a regular cadence.
For prototypes, side projects, and low-traffic tools, yes. Production traffic will usually hit a rate limit quickly. Treat the free tier as an evaluation channel: validate the model and provider, then move to a paid endpoint with the same model when you scale.
Either no provider currently offers it for free, the free promotion ended, or it has not been benchmarked yet. Open the model's main page to compare paid options, or let us know about a missing free provider via the feedback link in the footer.
