Models served via OpenRouter

15 models from 9 makers that you can call through OpenRouter — each one tested by us with fixed prompts, outputs kept, costs measured.

Makers: anthropic · google · meta · mistral · moonshotai · openai · qwen · xai · z-ai

15 models 🧪 all independently tested

gemini-3-pro-image

131K

gemini-2.5-flash-image

33K

gemini-3.1-flash-image

131K

gemma-3-4b-it

131K

Claude 3 Haiku

200K

Llama 4 Scout

10.0M

Mistral Small 3.2 24B

131K

GPT-5 Nano

400K

Grok Build 0.1

256K

Kimi K2.5

262K

GPT-5 Image

400K

GPT-5 Image Mini

400K

GPT-5.4 Image 2

272K

Qwen3.5-Flash

1.0M

Z.ai: GLM 4.6V

131K

About OpenRouter

aggregator

OpenRouter doesn't train models — it's an aggregator that sits in front of OpenAI, Anthropic, Google, Meta, DeepSeek, and dozens of other upstream providers behind a single API key and a single bill.

The practical appeal is less integration work: instead of juggling separate SDKs, keys, and invoices for every provider you want to try, you point at one endpoint and switch models by changing a string. Most routes bill at the upstream provider's own per-token rate, and OpenRouter will automatically fail over to a backup route if the primary provider is down or rate-limited.

The tradeoff is that you're one layer removed from the source: pricing, rate limits, and model availability all depend on OpenRouter's relationship with the upstream provider, not just the provider itself.

  • One API key routes to 400+ models from 70+ upstream providers
  • Most models bill at the upstream provider's own rate, not a markup
  • Automatic fallback routing when an upstream provider is down or rate-limited

Affiliate program: None found.

OpenRouter pricing, by model

This is what OpenRouter itself charges for each model — not the cheapest price across every provider (that's what the cards above show). Prompt tokens range $0.0000–$10.00 per 1M across 15 priced models.

ModelMakerPrice via OpenRouterContext
Anthropic: Claude 3 Haikuanthropic$0.25 / $1.25 per 1M200K
gemini-2.5-flash-imagegoogle$0.30 / $2.50 per 1M33K
gemini-3-pro-imagegoogle$2.00 / $12.00 per 1M66K
gemini-3.1-flash-imagegoogle$0.50 / $3.00 per 1M131K
gemma-3-4b-itgooglefree33K
Meta: Llama 4 Scoutmeta$0.10 / $0.30 per 1M10.0M
Mistral: Mistral Small 3.2 24Bmistralfree131K
MoonshotAI: Kimi K2.5moonshotai$0.35 / $1.89 per 1M262K
OpenAI: GPT-5 Imageopenai$10.00 / $10.00 per 1M400K
OpenAI: GPT-5 Image Miniopenai$2.50 / $2.00 per 1M400K
OpenAI: GPT-5 Nanoopenai$0.05 / $0.40 per 1M400K
OpenAI: GPT-5.4 Image 2openai$8.00 / $15.00 per 1M272K
Qwen: Qwen3.5-Flashqwen$0.07 / $0.26 per 1M1.0M
xAI: Grok Build 0.1xai$1.00 / $2.00 per 1M256K
Z.ai: GLM 4.6Vz-ai$0.30 / $0.90 per 1M131K

Latency: Per-provider latency isn't independently measured yet — the probe rig currently tests one route per model, not one per provider (see PROBE_RIG.md ledger #10).

FAQ

Is OpenRouter its own model, or does it host other providers' models?
OpenRouter doesn't train models — it's a unified API that proxies to upstream providers (OpenAI, Anthropic, Google, Meta, DeepSeek, and dozens more), so you get one API key and one bill instead of juggling a dozen.
Does OpenRouter mark up pricing?
Most routes bill at the upstream provider's own per-token rate; check the per-model pricing table above since a minority of routes carry a small markup for infrastructure.