Model catalog
Ten open-weight models, one for each job. Or let linkrra/auto choose for you.
Zero data retention| Model | Best for | Input / 1M | Output / 1M | Why it is on this list |
|---|---|---|---|---|
| linkrra/auto | Anything | varies | varies | Reads your request and picks the cheapest model below that can do it. You are billed at that model's rate, and the response tells you which one ran and why. |
| qwen3-coder | Coding, hard | $0.506 | $1.424 | The full coder, for agentic and long-horizon engineering work. |
| deepseek-v4-pro | Reasoning, deep | $2.279 | $4.051 | DeepSeek's flagship, for problems that are genuinely hard. |
| deepseek-v4-flash | Fast chat | $0.150 | $0.299 | Near-instant answers. Built for support widgets and high-volume agent loops. |
| llama-3.1-8b-instruct | Workhorse | $0.084 | $0.135 | The default. Fast, cheap, and the model most developers have already tested against. |
| glm-5.3 | Multilingual | $1.994 | $5.570 | The strongest open model outside English. Newest GLM release. |
| glm-5.3-flash | Multilingual, fast | $0.152 | $0.506 | GLM's multilingual strength at flash speed and price. |
| qwen3.8-27b | Balanced | $0.361 | $3.632 | Near-flagship Qwen quality with cheap input. A strong all-round default. |
| kimi-k3 | Long context | $3.772 | $16.82 | 975k context. An entire codebase or case file in a single call. |
| minimax-m2.5 | Agents | $0.456 | $1.538 | Made for long tool-calling chains, where a weaker model wastes more in retries than it saves. |
| gpt-oss-120b | Reasoning, cheap | $0.062 | $0.287 | Frontier-class reasoning at workhorse prices. Open weights, from OpenAI. |
| gemma-4-31b-it | Compact | $0.152 | $0.574 | Google's open Gemma — small, efficient, and genuinely open-weight. |
| qwen3-coder-30b-a3b-instruct | Coding, everyday | $0.118 | $0.473 | Purpose-built for code. Handles most completion and refactor work. |
| qwen3-vl-30b-a3b-instruct | Vision | $0.253 | $0.855 | Reads images, screenshots and scanned documents. |
| gemini-3.1-flash-lite-image | Image | $0.317 | $1.899 | Fastest image generation — the cheapest way to a picture. |
| gemini-3.1-flash-image | Image | $0.633 | $3.798 | Fast, high-quality image generation and editing. |
| gemini-3-pro-image | Image | $2.532 | $15.19 | Google's best image model — photorealistic output, precise edits. |
| gemini-2.5-flash-image | Image | $0.380 | $3.165 | The original flash image model — solid quality, low price. |
| gpt-5.4-image-2 | Image | $10.13 | $18.99 | OpenAI's newest image model — strongest prompt adherence. |
| gpt-5-image | Image | $12.66 | $12.66 | OpenAI's flagship image model. |
| gpt-5-image-mini | Image | $3.165 | $2.532 | Cheaper OpenAI image generation. |
| gpt-audio | Voice | $3.165 | $12.66 | Voice-capable chat — understands speech and can speak its answers. |
| gpt-audio-mini | Voice | $0.760 | $3.038 | The lighter voice model — the engine behind spoken answers. |
Image & voice models
Added 2026-09-01. Image models generate pictures inside the chat; voice models speak their answers. No open-weight supply exists for these modalities on OpenRouter, so they are closed-weight by documented exception (note below).
| Model | Type | Input / 1M | Output / 1M | What it is for |
|---|---|---|---|---|
| gemini-3.1-flash-lite-image | Image | $0.300 | $1.800 | Fastest image generation — the cheapest way to a picture. |
| gemini-3.1-flash-image | Image | $0.600 | $3.600 | Fast, high-quality image generation and editing. |
| gemini-3-pro-image | Image | $2.400 | $14.400 | Google's best image model — photorealistic output, precise edits. |
| gemini-2.5-flash-image | Image | $0.360 | $3.000 | The original flash image model — solid quality, low price. |
| gpt-5.4-image-2 | Image | $9.600 | $18.000 | OpenAI's newest image model — strongest prompt adherence. |
| gpt-5-image | Image | $12.000 | $12.000 | OpenAI's flagship image model. |
| gpt-5-image-mini | Image | $3.000 | $2.400 | Cheaper OpenAI image generation. |
| gpt-audio | Voice | $3.000 | $12.000 | Voice-capable chat — understands speech and can speak its answers. |
| gpt-audio-mini | Voice | $0.720 | $2.880 | The lighter voice model — the engine behind spoken answers. |
Prices per 1M tokens. Input and output are billed separately, at the rates shown —
most requests are mostly input, so what you pay is usually much closer to the left-hand
column. Generated from the live catalogue on 2026-09-18 from the same file the
gateway bills from, so this page cannot quote a price we do not charge.
Open weights only. Every model here has publicly downloadable weights,
which is what lets us move it onto our own hardware later without you changing a line.
191 proprietary models are excluded on
purpose.
Zero data retention, enforced per request. Every call we forward
carries zdr: true and data_collection: deny. A model with no
compliant endpoint fails the request rather than quietly routing to a provider that
keeps your data — one model was dropped from this list for exactly that reason.