Mistral AI

Executive Summary

What it is: Mistral AI offers a coding agent (Mistral Vibe) via CLI and IDE, alongside a web chat (Le Chat) and API access to its model lineup (Mistral Medium 3.5, Mistral Large 3, Mistral Small 4, Codestral, Devstral 2). Consumer plans range from $0 (Free) to $24.99/user/mo (Team), with Enterprise pricing custom. Pro at $14.99/mo includes the Vibe CLI, priced below Anthropic Pro ($20/mo), OpenAI Plus ($20/mo), and Cursor Pro ($20/mo). API pricing for coding-relevant models ranges from $0.15/$0.60 per MTok (Mistral Small 4) to $1.50/$7.50 per MTok (Mistral Medium 3.5).

What to watch out for: All usage limits on the pricing page are described as multiples of an undisclosed Free baseline ("Up to 6x Free," "Up to 30x Free") with no concrete numbers published. Mistral Vibe for IDE is Enterprise-only. Batch API offers 50% discount on all models.

Bottom line: Mistral offers a competitively priced Pro tier ($14.99/mo) and its open-weight models (Devstral 2, Mistral Large 3) enable self-hosting, one of the few suppliers in this report where you can run coding agents on your own infrastructure without a commercial license. However, the complete absence of concrete usage numbers makes cost planning impossible without a trial. Best suited for teams that value open-weight models and self-hosting options over a polished, fully managed agent experience.

Key Terms

  • Mistral Vibe - Mistral AI's coding agent, available as a CLI tool and IDE integration. Included with Le Chat Pro and Team plans, with pay-as-you-go billing beyond the included usage. Source: Mistral AI – Vibe
  • Codestral - Mistral's specialist code completion model (v25.08), available via API. Designed for FIM (Fill-in-the-Middle) completion and code generation. Source: Mistral AI – Codestral 25 08
  • Devstral 2 - Mistral's open-weight code agent model (v25.12) for solving software engineering tasks. Suitable for agentic coding workflows. Source: Mistral AI – Devstral 2 25 12
  • Le Chat - Mistral's web-based AI chat interface, similar to ChatGPT. Available in Free, Pro, Team, and Enterprise tiers. Source: Mistral AI – Pricing
  • Premier model - a Mistral model designation for proprietary, closed-weight models that are only available through Mistral's own API and Le Chat (e.g., Mistral Medium 3.1, Codestral). Source: Mistral AI – Overview
  • Open model - a Mistral model with open weights that can be self-hosted (e.g., Mistral Large 3, Mistral Small 4, Devstral 2). Source: Mistral AI – Overview

Latest Changes

First report for this supplier. All models, plans, and pricing are listed as current state.

  • Feature added: Workflows launched in public preview (April 27). Automated AI-driven business processes.
  • New model: Mistral Small 4 released March 2026. Open-weight hybrid model unifying instruct, reasoning, and coding.
  • Feature added: Mistral Vibe CLI included with Pro and Team plans.
  • Deprecation (upcoming): Mistral Large 2.1 and Pixtral Large retire May 31, 2026. Replace with Mistral Large 3.
  • Deprecation: Devstral Small 2 retired March 31, 2026. Replace with Devstral 2.

Plans

PlanPriceKey Inclusions
Free$0Personal AI assistant, Le Chat, SOTA models, 500 memories, 40+ enterprise connectors
Pro$14.99/month (excluding taxes)Everything in Free + up to 6x free messages/web searches, up to 30x free think mode, up to 5x free deep research, 15GB storage, Mistral Vibe CLI (PAYG beyond), image generation
Team$24.99/user/month (excluding taxes)Everything in Pro + 30GB/user storage, domain verification, data export, admin API
EnterpriseCustom (contact sales)Everything in Team + SAML SSO, audit logs, white label, Mistral Vibe for IDE, async coding agents, custom MCP connectors

Source: Mistral AI – Pricing

Feature Comparison (Mistral's Published Table)

FeatureFreeProTeamEnterprise
MessagesBaselineUp to 6x freeUp to 6x freeCustom
Web searchesBaselineUp to 6x freeUp to 6x freeCustom
Think modeBaselineUp to 30x freeUp to 30x freeCustom
Deep researchBaselineUp to 5x freeUp to 5x freeCustom
Memories5001,0001,0001,000
Document storageLimitedUp to 15GBUp to 30GB/userCustom
Mistral Vibe CLINot includedIncluded (PAYG beyond)Included (PAYG beyond)Custom
Mistral Vibe for IDENot includedNot includedNot includedCustom
Async coding agentsNot includedNot includedNot includedCustom
Student discountN/AUndiscounted price undisclosedN/AN/A

Source: Mistral AI – Pricing

API Pricing

ModelTypeInput ($/MTok)Output ($/MTok)Batch Discount
Mistral Medium 3.5Premier$1.50$7.5050%
Mistral Large 3Open$0.50$1.5050%
Mistral Small 4Open$0.15$0.6050%
Devstral 2Open$0.40$2.0050%
CodestralPremier$0.30$0.9050%
Magistral MediumPremier$2.00$5.0050%
Magistral SmallPremier$0.50$1.5050%
Mistral Small 3.2Open$0.10$0.3050%
Devstral Small 2Open (Labs)$0.10$0.3050%

Fine-tuning available for: Mistral Small 3.2 ($4/MTok training, $2/mo storage), Pixtral 12B ($2/MTok training), Ministral models ($1/MTok training).

Other API pricing:

  • OCR 3: $2/1,000 pages (50% off with batch API)
  • Code execution: $30/1,000 calls
  • Web search: $30/1,000 calls
  • Image generation: $100/1,000 images
  • Agent API: model cost per MTok + tool call

Enterprise APIs (Priority Tier): Priced starting at 75% above list pricing. Includes regional data processing controls, system-level SLAs, and increased rate limits.

Source: Mistral AI – Pricing

Model Performance / Benchmarks

Mistral does not publish independent benchmark scores for Mistral Vibe as a product. Benchmark scores for individual Mistral models (Mistral Medium 3.5, Mistral Large 3, Small 4, Codestral, Devstral 2) are not documented on the pricing or model pages.

Mistral Medium 3.5 is described as "SOTA in long-horizon instruction-following, reasoning, and coding." Devstral 2 is described as "state-of-the-art, open-weights agentic coding model for autonomous software engineering." No quantitative benchmark data accompanies these claims.

Latest News

Workflows Launch (April 27, 2026)

Mistral launched Workflows in public preview, a product for building automated AI-driven business processes. This is not directly related to coding agents but expands the Mistral platform for enterprise customers.

Source: Mistral AI – Workflows

Mistral Small 4 Release (March 2026)

Mistral Small 4 was released as a hybrid model unifying instruct, reasoning, and coding in a single efficient model. It is open-weight and suitable for self-deployment.

Source: Mistral AI – Mistral Small 4 0 26 03

Model Deprecations (Ongoing)

Multiple legacy models are scheduled for deprecation:

  • Mistral Large 2.1 (v24.11): Deprecation February 27, 2026, Retirement May 31, 2026. Replacement: Mistral Large 3.
  • Pixtral Large (v24.11): Deprecation February 27, 2026, Retirement May 31, 2026. Replacement: Mistral Large 3.
  • Devstral Small 2 (labs): Deprecation February 27, 2026, Retirement March 31, 2026. Replacement: Devstral 2.
  • Codestral (v25.01): Deprecation November 6, 2025, Retirement November 30, 2025. Replacement: Codestral (v25.08).

Source: Mistral AI – Overview

Community Signals

Mistral Vibe CLI Reception

Mistral Vibe CLI is new and community feedback is still emerging. Users on Reddit report it as a promising but early-stage coding agent compared to Claude Code and Cursor.

Open-Weight Model Community

Devstral 2 (open-weight code agent model) has received positive attention for being freely available for self-hosting, making it attractive for organizations that want to run coding agents on their own infrastructure.

Pricing Transparency Criticism

Mistral's pricing transparency has been criticized in the community. The "Up to 6x free" and "Up to 30x Free" usage descriptions on the pricing page do not provide concrete numbers, making it impossible to compare actual usage limits with competitors.

Model Lineup Complexity

The model lineup is complex, with many overlapping models (Mistral Small 3.2, Small 4, Ministral 3 14B, etc.) and a confusing naming scheme that makes model selection difficult for new users.

Enterprise Readiness

FeatureAvailable?Details
SSO (SAML)YesEnterprise plan. Source: Mistral AI – Pricing
SSO (OIDC)UndisclosedNot explicitly mentioned.
SCIMNoNot mentioned on pricing page.
Audit logsYesEnterprise plan. Source: Mistral AI – Pricing
IP indemnityNoNot mentioned on pricing or product pages.
Data residencyNoNot mentioned on pricing page.
HIPAANoNot mentioned on pricing page.
Air-gapped / on-premPartialOpen-weight models (Devstral 2, Mistral Large 3) can be self-hosted on your own infrastructure. Managed platform is cloud-only. Source: Mistral AI – Overview
SLANoNo published SLA on pricing page.
Admin controls (RBAC)YesWhite label, domain verification, data export, admin API on Team and Enterprise. Source: Mistral AI – Pricing

Transparency Gaps

MetricStatusNotes
Direct API per-token pricingDisclosedMistral Medium 3.5: $1.50/$7.50, Mistral Large 3: $0.50/$1.50, Devstral 2: $0.40/$2.00, Codestral: $0.30/$0.90 per MTok. Source: Mistral AI – Pricing
Le Chat message limitsundisclosedDescribed as "Up to 6x free" without baseline numbers
Web search limitsundisclosedDescribed as "Up to 6x free" without baseline numbers
Think mode limitsundisclosedDescribed as "Up to 30x free" without baseline numbers
Deep research limitsundisclosedDescribed as "Up to 5x free" without baseline numbers
Mistral Vibe PAYG pricingundisclosed"PAYG beyond" included usage, but per-token rate not published
Free tier baselineundisclosedNo concrete numbers for any usage metric