mistral-small-4 API — compare providers

The same model (mistral-small-4) is served by 2 providers; Mistral AI is the cheapest at $0.75/1M blended — the priciest (Venice) costs 25% more.

RankProviderInput /1MOutput /1MBlended /1MSpeed (tok/s per $)
1Mistral AI$0.15$0.6$0.75 ← cheapest195.3
2Venice$0.1875$0.75$0.938

Speed per dollar = measured output tokens/s ÷ blended price. Throughput is provider-specific, so it can rank endpoints differently from price — see the speed-per-dollar ranking.

Prices are per 1M tokens (USD). "Blended" = input + output, for coarse ranking. Same underlying model, 2 serving providers, 25% spread top to bottom. Context window served: 262,144 tokens (32,768 max output) on all 2 priced endpoints — see how it ranks in biggest context windows. See how this compares across the catalogue in same model, different price.

mistral-small-4 spec sheetmaker-declared

What Mistral AI declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.

Context window
262,144 tokens
Parameters
119,000,000,000≈ 119.0B
Active experts
4 of 128 active per tokenmaker's docs · read 2026-09-07
Weight precision
to verify
License
Apache 2.0 — open weights, permissivecurated · read 2026-09-07Scope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
Knowledge cutoff
not published by the maker
Modalities
text · image → textmaker's docs · read 2026-09-07
Max output (official)
not published by the maker
Release date
March 16, 2026
Maker lifecycle status
General availability

docs.mistral.ai · maker's documentation · read 2026-09-07 · 7 of 10 fields published by the maker

Speed by provider measured

Identical weights, different machines: the endpoint decides how fast the model answers. Our own measurements — a fixed prompt, streamed, 256 output tokens — taken from eu-paris and us-east.

ProviderBlended /1Meu-parisus-east
TTFT p50tok/sTTFT p50tok/s
Mistral AI$0.75216 ms147340 ms147

TTFT = time to first token, p50 over 12 samples per endpoint and per region; tok/s = output tokens per second while streaming. Last measured: eu-paris 2026-09-06, us-east 2026-09-06. 1 of the 2 providers serving mistral-small-4 is in the bench — the others have no figure here, which says nothing about their speed. Latency moves with load and route: read it as an order of magnitude, not a guarantee. See the measurement protocol and the cross-catalogue speed-per-dollar ranking.

Cost per 1,000 requests by workload

What each provider actually bills for a representative job, not just the sticker price.

ProviderChatbot
1000 in / 500 out
RAG / long context
8000 in / 500 out
Batch summarize
4000 in / 1000 out
Mistral AI$0.45$1.50$1.20
Venice$0.56$1.88$1.50

Estimate your own workload

Input tokens/request:   Output tokens/request:   Requests:

ProviderEstimated cost (USD)

Self-host / quantized builds

Open weights under Apache 2.0 — you can run mistral-small-4 on your own hardware instead of paying per token. Even quantized to int4 it needs roughly 71.4 GB of VRAM — a multi-GPU workstation or a large-memory Mac, not a typical desktop. Formats: GGUF for llama.cpp / Ollama / LM Studio, AWQ and GPTQ for GPU serving, MLX for Apple silicon.

Links search the live model hubs (Hugging Face, Ollama) so they track new builds as they appear — we don't host weights. Which formats exist depends on what the model owner and community have published.

→ See VRAM, GPU and cost vs API for mistral-small-4, or the full self-host break-even calculator.