mistral-large-4-0
Auto-routermistral/mistral-large-4-0
Context
1M
Max output
—
Input / 1M
$1.36
Output / 1M
$4.18
Cached / 1M
$0.14
Cutoff
—
About this model
Mistral Large 4 (v26.10) is a state-of-the-art, natively multimodal general-purpose model with a granular Mixture-of-Experts architecture, featuring 52B active parameters and 1.05T total parameters plus a 1.6B vision encoder. Released 2026-10-06; currently in Public Preview. 1M token context window.
Best suited for
- Complex reasoning and enterprise tasks
- General-purpose multimodal tasks
- Long-context document analysis (1M tokens)
- Coding and software development
- Function calling and agentic workflows
- Structured output generation
Capabilities
Vision
Accepts images alongside text in the same message.
Files
Accepts file attachments — PDFs, transcripts, spreadsheets.
Tools
Native function calling, so agents can invoke your endpoints.
System prompt
Honours a dedicated system role, separate from the user turn.
Supported parameters
additional_parametersUse this model
OpenAI-SDK compatible, with gateway fallback and routing across providers.
Other Mistral models
| Model | Context | Max output | Input | Output | Capabilities |
|---|---|---|---|---|---|
| zai-glm-5-3 Chat | 1M | 128K | $1.40 | $4.40 | Tools, System prompt |
| ministral-8b-2512 Chat | 256K | 33K | $0.15 | $0.15 | Vision, Files, Tools, System prompt |
| mistral-large-2512 Chat | 256K | 131K | $0.50 | $1.50 | Vision, Files, Tools, System prompt |
| mistral-small-latest Chat | 128K | 128K | $0.15 | $0.60 | Vision, Tools, System prompt |
| codestral-latest Chat | 256K | 40K | $0.30 | $0.90 | Vision, Tools, System prompt |
| mistral-medium-latest Chat | 128K | 128K | $1.50 | $7.50 | Vision, Tools, System prompt |
Pricing & provider details
Mistral
mistral/mistral-large-4-0
- Input · per 1M
- $1.36
- Output · per 1M
- $4.18
- Cached input · per 1M
- $0.14
- Context window
- 1,000,000 tokens
- Max output
- —
- Knowledge cutoff
- —
- Auto-router
- Supported
Start building today
Route mistral-large-4-0 — and every other model in the catalogue — through one endpoint, with failover built in.