Free LLM APIs (2026)

Free tier: Free tier (no billing) on Flash/Flash-Lite models incl. gemini-3.8-flash, gemini-3.6-flash, gemini-3.5-flash-lite, gemini-2.5-pro/flash/flash-lite, plus Flash TTS: input/output "Free of charge". Not free: Gemini 4 Argon, 3.1 Pro preview, image models. Per-model RPM/RPD for the free tier are not published on the page ("can be viewed in Google AI Studio"); RPD resets at midnight Pacific.

• Auth: API key (x-goog-api-key)
• Serving public users: Allowed, but terms say: "You may use only Paid Services when making API Clients available to users in the European Economic Area, Switzerland, or the United Kingdom."
• Note: Free-tier ("Unpaid Services") prompts and outputs are used by Google to improve products; don't send sensitive data.
• Source: ai.google.dev · ai.google.dev · ai.google.dev

Free tier: Free plan per-model limits, e.g. openai/gpt-oss-120b and gpt-oss-20b: 30 RPM, 1K RPD, 8K TPM, 200K TPD; qwen/qwen3.8-27b: 30 RPM, 1K RPD, 8K TPM, 200K TPD; whisper-large-v3(-turbo): 20 RPM, 2K RPD, 7.2K audio-sec/hour, 28.8K audio-sec/day; Orpheus TTS: 10 RPM, 100 RPD.

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Services agreement grants the right to make the services available to End Users through your Customer Applications; no reselling of account access.
• Note: Page says limits are a "high level summary" and exact org limits are on the console Limits page.
• Source: console.groq.com · console.groq.com

Free tier: Models with IDs ending in ":free": 20 requests/min; 50 requests/day if you have purchased <10 credits all-time, 1000 requests/day if you have purchased ≥10 credits. FAQ: "usually not suitable for production use".

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Terms don't explicitly cover serving public end users; each underlying provider's terms apply. Free model providers may log prompts.
• Note: A negative credit balance can block free models too.
• Source: openrouter.ai · openrouter.ai · openrouter.ai

Free tier: "Free mode": API access enabled by default, no credit card; "usage and rate limits apply". Exact free limits (requests/sec, tokens/min, tokens/month per model) are not published publicly, only on Admin Panel › API › Limits. Exact limits: unknown.

• Auth: API key (Bearer)
• Serving public users: Unknown for production; free mode is positioned for evaluation/prototyping.
• Note: Mistral help center describes Free mode as intended for evaluation and prototyping.
• Source: docs.mistral.ai · docs.mistral.ai

Free tier: 10,000 Neurons per day free on both Workers Free and Paid plans (resets 00:00 UTC); beyond that requires Workers Paid at $0.011 per 1,000 Neurons. Some models require a paid billing method.

• Auth: Cloudflare API token (Bearer) or Workers binding
• Serving public users: Yes (standard Cloudflare service; built for production apps).
• Note: Covers text, image (e.g. FLUX/SDXL), speech and embedding models in one quota.
• Source: developers.cloudflare.com

Free tier: "All models on build.nvidia.com are free to prototype with"; free tier is rate-limited "up to 40 requests per minute (RPM) for most models, with no per-token billing".

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Prototyping only per FAQ wording.
• Note: Positioned for prototyping; production means partner endpoints or self-hosted NIM.
• Source: build.nvidia.com

Free tier: Trial (evaluation) keys: 1,000 API calls per month; per-endpoint trial rate limits (e.g. Command A+ chat 20 req/min, audio transcription 5 req/min).

• Auth: API key (Bearer)
• Serving public users: Trial keys are described as "evaluation keys"; production use needs a paid production key.
• Note: Includes Embed and Rerank endpoints under the same trial key.
• Source: docs.cohere.com

Free tier: Not permanently free: new accounts get $5 in credits after adding a verified payment method, expiring 30 days after grant; per-model Free Trial rate limits apply.

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Trial only.
• Note: FAQ: "Is there a permanently free tier? No."
• Source: inference-docs.cerebras.ai

Free tier: Every Vercel team gets $5/month in AI Gateway credits on "Free Tier eligible models" (a subset); free-tier requests are rate-limited per model, lower than paid. Credits start at first request.

• Auth: API key or Vercel OIDC token, OpenAI-compatible
• Serving public users: Yes (Vercel product), within free-tier limits.
• Note: Paid tier: provider list price, zero markup.
• Source: vercel.com

Free tier: GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash listed as "Free" (input, cached input, output). Several larger GLM models show "Limited-time Free" only for cached-input storage. Rate limits for free models: unknown.

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Unknown.
• Source: docs.z.ai

Free tier: Free tier: 1 million tokens and 60 minutes of audio transcription (to try models); some models (e.g. lightonocr) listed as Free.

• Auth: API key (Bearer), OpenAI-compatible
• Serving public users: Yes (commercial cloud service).
• Note: EU-hosted (France). Period of the 1M-token free tier (one-time vs monthly) not stated: unknown.
• Source: www.scaleway.com

Free tier: Free plan: run models locally plus "starter usage credits" for cloud models and access to starter models. Credit amount and limits: unknown (not stated).

• Auth: API key (Bearer)
• Serving public users: Unknown.
• Source: ollama.com