Mistral La Plateforme
Mistral AI · Ranked #4 of 7 in LLM APIs
European open-weight-friendly provider with a genuine experiment free tier and competitive flagship pricing.
Open-weight European LLMs

Overview
Mistral La Plateforme is the first-party developer API for Mistral AI, the French frontier-lab founded in 2023. It exposes Mistral's full model portfolio behind a single OpenAI-compatible REST surface: general-purpose text models (Mistral Large 3, Medium 3.5, Small 4, the Ministral 3 edge family), reasoning models (Magistral), code models (Codestral, Devstral 2), plus multimodal and document tooling (Pixtral vision, OCR 4, Voxtral audio/TTS), embeddings, moderation, fine-tuning and an Agents API with native MCP support. Its distinguishing pitch is European data residency, openness (many models ship with open weights you can self-host), and aggressive price-performance, Mistral Large 3 lists at just $0.50/$1.50 per 1M input/output tokens, far below comparable US flagships.
For developers, the platform is genuinely easy to adopt: a generous free "Experiment" tier (roughly 1B tokens/month, rate-limited, for evaluation only), official Python and TypeScript SDKs, JSON mode and structured outputs, function calling, and a 50% batch discount for async jobs. Third-party benchmarks from Artificial Analysis show Mistral's own endpoint is strong on latency (it posts among the lowest time-to-first-token of any provider serving its models) but middling on raw throughput, for Mistral Large 3, Mistral's endpoint delivers ~49 tok/s versus ~215 tok/s from Amazon Bedrock serving the same open-weight model. This is the recurring trade-off: because Mistral open-sources many models, cloud partners (Bedrock, Azure, Vertex) often out-serve the first-party API on throughput, while Mistral wins on latency, price, and feature freshness.
The main weaknesses are operational and reputational rather than technical. Trustpilot and forum reports cite intermittent downtime, slow/templated support, and rough edges on newer agentic features, and several developers describe model quality (especially Devstral via API) as inconsistent relative to marketing. Aggregate review volume is thin, G2 shows only a handful of reviews, so sentiment signal is noisy. Net: La Plateforme is a compelling choice for cost-sensitive, EU-compliance-driven, or open-weight-friendly teams who value latency and a clean OpenAI-compatible API, but throughput-critical or mission-critical production workloads may prefer running Mistral's open models on a hyperscaler.
How this score is derived
The APIbenchmarks Index is a weighted sum of four dimensions, each scored on an absolute 0–100 reference scale. See the methodology for every mapping.
| Dimension | Score | Weight | Contribution |
|---|---|---|---|
| Documentation & DXDocs at docs.mistral.ai are well-organized with model overviews, capability guides, rate-limit/tier tables and quickstarts, though some users report support replies with inaccurate AI-generated doc references. | 82 | 30% | 24.6 |
| ReliabilityMistral's first-party endpoint posts low time-to-first-token in Artificial Analysis benchmarks, but Trustpilot users report intermittent downtime and slow/failed requests, and there is no prominently published uptime SLA. | 74 | 25% | 18.5 |
| Ecosystem & SDKsStrong ecosystem via OpenAI-compatible API, official Python/TS SDKs, MCP-native Agents API, and wide availability of its open-weight models on Bedrock, Azure and Vertex. | 78 | 25% | 19.5 |
| AccessibilityVery accessible: a free ~1B-token/month Experiment tier, pay-as-you-go auto-upgrade, and some of the lowest per-token prices among frontier labs lower the barrier to entry. | 88 | 20% | 17.6 |
| APIbenchmarks Index (ABI) | 80.2 | ||
Table 1. Derivation of the ABI for Mistral La Plateforme. Contribution = score × weight; the index is their sum.
At a glance
- Vendor
- Mistral AI
- Pricing model
- Per token
- Free tier
- Free Experiment tier, rate-limited (~1B tokens/mo cap), all models
- Official SDKs
- 6 languages
Pricing
| Mistral Large 3 (mistral-large-latest) | $0.50 / $1.50 per 1M tokens | Flagship general model, input/output |
| Mistral Medium 3.5 (mistral-medium-latest) | $1.50 / $7.50 per 1M tokens | Highest-tier multimodal model |
| Mistral Small 4 (mistral-small-latest) | $0.15 / $0.60 per 1M tokens | Efficient mid-size model |
| Ministral 3 - 3B (ministral-3b-latest) | $0.10 / $0.10 per 1M tokens | Cheapest edge/small model |
| Codestral (codestral-latest) | $0.30 / $0.90 per 1M tokens | Code-completion model |
| Free Experiment tier + Batch API | $0 free tier; 50% off batch | ~1B tokens/month rate-limited eval tier; 50% batch discount on all production rates (24h SLA) |
Key features
- •OpenAI-compatible chat completions REST API
- •JSON mode and structured outputs
- •Function / tool calling
- •Agents API with native MCP support
- •Vision / multimodal (Pixtral, Medium 3.5)
- •OCR 4 document AI with bounding boxes
- •Voxtral audio transcription and TTS with voice cloning
- •Embeddings (Mistral Embed, Codestral Embed)
- •Fine-tuning and content moderation
- •50% batch API discount for async jobs
Official SDKs
Strengths & trade-offs
- +Aggressively low pricing, Mistral Large 3 at $0.50/$1.50 per 1M tokens undercuts comparable US frontier flagships
- +Many models ship as open weights, enabling self-hosting and avoiding vendor lock-in
- +Lowest time-to-first-token among providers serving its models in Artificial Analysis benchmarks
- +Generous free Experiment tier (~1B tokens/month) plus 50% batch discount for async workloads
- +OpenAI-compatible API with official Python and TypeScript SDKs and MCP-native Agents API
- +European (EU) data residency appealing for GDPR/compliance-sensitive teams
- –First-party endpoint throughput trails cloud partners, ~49 tok/s on Large 3 vs ~215 tok/s on Amazon Bedrock
- –Reports of intermittent downtime and slow/failed requests on Trustpilot
- –Support described as slow and sometimes templated/AI-generated with inaccurate references
- –Newer agentic and Devstral capabilities described by some developers as underwhelming via API
- –No prominently published uptime SLA on the public site
- –Thin third-party review volume (only a handful of G2 reviews) makes reliability hard to assess
What developers say
G2 4.5/5 · 3 reviews
Developers praise Mistral's low cost, open weights, easy OpenAI-compatible integration and friendly MCP support, but criticize inconsistent model quality, intermittent downtime and slow support.
“The ability to enrich either a chat or generally an agent with one or more libraries has been solved in a very friendly way, more accessible than comparable offerings from OpenAI or Anthropic.”
Key figures
| Output speed, Mistral Large 3 (Mistral endpoint) | 49.4 tokens/sec | Artificial Analysis ↗ |
| Time to first token, Mistral Large 3 (Mistral endpoint) | 1.20 s | Artificial Analysis ↗ |
| Blended price, Mistral Large 3 (Mistral endpoint) | $0.60 per 1M tokens (7:2:1 blend) | Artificial Analysis ↗ |
| Fastest output speed across Mistral models, Mistral Small 4 | 168.0 tokens/sec | Artificial Analysis ↗ |
| Lowest latency across Mistral models, Ministral 3 3B | 0.69 s time to first token | Artificial Analysis ↗ |
| List price, Mistral Large 3 | $0.50 input / $1.50 output per 1M tokens | Mistral pricing page ↗ |
| Batch API discount | 50% off standard rates | Mistral pricing page ↗ |
Compare Mistral La Plateforme head to head
Sources
- https://mistral.ai/pricing/
- https://docs.mistral.ai/getting-started/models/models_overview/
- https://docs.mistral.ai/getting-started/clients
- https://artificialanalysis.ai/providers/mistral
- https://artificialanalysis.ai/models/mistral-large-3/providers
- https://github.com/mistralai/client-python
- https://www.g2.com/products/mistral-ai/reviews
- https://www.trustpilot.com/review/mistral.ai
- https://news.ycombinator.com/item?id=44107187
Figures last verified 2026-06-27. Spotted an error? corrections@apibenchmarks.com
