Home / Models / Mistral Small 2603
Mistral

Mistral Small 2603

mistral/mistral-small-2603

Mistral Small — cheap, fast, JSON-mode native. Used in chat and extraction fallback chains.

Pricing

DirectionUSD per 1M tokensUSD per 1K tokens
Input$0.2000$0.000200
Output$0.6000$0.000600

Pass-through provider rates via GammaInfra. No per-token markup. A 3% top-up fee (launch window through 2026-06-23, then 5%) applies on managed credits; the BYOK alternative is 1–2% per request.

Capabilities

✓ Tool calling — Vision (multimodal) ✓ JSON mode

Specifications

FieldValue
Context window128K
Max output32K
Providermistral
StreamingYes — OpenAI-compatible SSE

Best for

chat extraction

Task labels reflect where this model heads or appears in GammaInfra's default dispatcher chains. Override per-request with the X-GammaInfra-Preference or X-GammaInfra-Cost-Quality header.

How to call it

Through GammaInfra's smart router with one of your GammaInfra API keys:

curl https://api.gammainfra.com/v1/chat/completions \
  -H "Authorization: Bearer sk-gammainfra-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mistral/mistral-small-2603",
    "messages": [
      {"role": "user", "content": "Hello, Mistral Small 2603!"}
    ]
  }'

Or via any OpenAI SDK — see the integrations page for setup with Cursor, Cline, LangChain, the Vercel AI SDK, and others.

Smart routing — or pin this model

You can call mistral/mistral-small-2603 directly (as above), or let GammaInfra's router pick the best-fit model per prompt. Use gammainfra/auto as the model name for task-aware routing, gammainfra/fast for latency-optimized hedged requests, or gammainfra/cheap for cost-optimized routing. The router considers task type, latency, and your X-GammaInfra-Cost-Quality dial when picking.

Related models

GPT-5 mini → $0.25/$2.00 per 1M
Llama 3.1 8B Instant (via Groq) → $0.05/$0.08 per 1M
Mistral Large 2512 → $2.00/$6.00 per 1M

Ready to try it?

Get a GammaInfra API key →

$3 free trial credit on signup, $10 minimum top-up. Pass-through provider token rates plus 3% top-up fee during the launch window (5% after 2026-06-23).

Frequently asked questions

How much does Mistral Small 2603 cost through GammaInfra?
Mistral Small 2603 (mistral/mistral-small-2603) is billed at the Mistral pass-through rate — $0.2 per 1M input tokens and $0.6 per 1M output tokens, with 0% token markup. GammaInfra's fee is taken at top-up time (3% during the launch window through 2026-06-23, 5% after), not per token; the BYOK option is 1–2% per request instead. Every response returns X-GammaInfra-Cost-USD with the exact spend for that call.
What is Mistral Small 2603's context window?
Mistral Small 2603 accepts up to 128K tokens of input context and returns up to 32K output tokens per request. GammaInfra passes the full window through with no truncation. Check this model's Specifications and notes above for any provider long-context surcharge on very large prompts.
Does Mistral Small 2603 support tool calling, vision, and JSON mode?
Tool / function calling: Yes. Vision (image input): No. Native JSON / structured-output mode: Yes. When tool calling is used, GammaInfra translates tool-call IDs across providers so OpenAI-shaped agent code keeps working regardless of which provider serves the request.
How do I call Mistral Small 2603 through GammaInfra?
Point any OpenAI-compatible SDK at https://api.gammainfra.com/v1 with your sk-gammainfra-... key, then set the model to mistral/mistral-small-2603 to pin Mistral Small 2603 directly — or use gammainfra/auto to let the smart router pick it when it is the best fit. Only base_url and api_key change; the rest of your OpenAI SDK code is unchanged.
When does GammaInfra's router pick Mistral Small 2603?
Mistral Small 2603 heads or appears in GammaInfra's chat, extraction dispatcher chains. With gammainfra/auto the router selects it when a prompt classifies into one of those task types and your cost/quality preference fits; pin mistral/mistral-small-2603 to force it regardless of routing. The X-GammaInfra-Endpoint response header always reports which model actually served the request.