Mistral Medium 3.5

Name: Mistral Medium 3.5
Author: Mistral AI

Multimodal

Mistral AI

Mistral Medium 3.5 is Mistral's 128B open-weight reasoning model released in April 2026 under the Modified MIT License. It excels at constrained generation (COLLIE) and agentic telecom support tasks, with a 256K token context window.

Key Specifications

Parameters

128.0B

Context

256.0K

Release Date

April 29, 2026

Average Score

66.3%

API Documentation Results Blog

Timeline

Key dates in the model's history

Announcement

April 29, 2026

Last Update

August 27, 2026

Today

September 10, 2026

Technical Specifications

Parameters

128.0B

Training Tokens

Knowledge Cutoff

Family

Capabilities

MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)

$1.50

Output (per 1M tokens)

$7.50

Max Input Tokens

256.0K

Max Output Tokens

256.0K

Supported Features

Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Programming

Programming skills tests

SWE-Bench Verified

SWE-Bench Verified. reasoning_effort="high". • Self-reported

78.0%

Other Tests

Specialized benchmarks

COLLIE

Collie. reasoning_effort="high". • Self-reported

96.0%

Tau3 Telecom

τ³-Telecom. reasoning_effort="high". User simulator: gpt-5.2 (reasoning_effort=low). 4 trials. • Self-reported

91.0%

AIME 2025

AIME 2025 avg@16. reasoning_effort="high". • Self-reported

86.0%

Beyond AIME

Beyond AIME avg@16. reasoning_effort="high". • Self-reported

66.9%

BrowseComp

BrowseComp. reasoning_effort="high". Context management with discard-all strategy at 100k tokens. • Self-reported

48.6%

Finance Agent v2

• Verified

32.1%

IFBench

AllenAI IFBench. reasoning_effort="high". • Self-reported

69.0%

Tau3 Airline

τ³-Airline. reasoning_effort="high". User simulator: gpt-5.2 (reasoning_effort=low). 4 trials. • Self-reported

72.0%

Tau3 Banking

τ³-Banking. reasoning_effort="high". User simulator: gpt-5.2 (reasoning_effort=low). Best of terminal- or embedding-based agentic search retrieval. • Self-reported

13.4%

Tau3 Retail

τ³-Retail. reasoning_effort="high". User simulator: gpt-5.2 (reasoning_effort=low). 4 trials. • Self-reported

76.1%

License & Metadata

License

modified_mit_license

Announcement Date

April 29, 2026

Last Updated

August 27, 2026

Similar Models

All Models

Pixtral Large

Mistral AI

MM124.0B

Released:Nov 2024

Price:$2.00/1M tokens

Mistral Large 3 (675B Instruct 2512)

Mistral AI

MM675.0B

Best score:0.4 (GPQA)

Released:Dec 2025

Price:$0.50/1M tokens

Mistral Small 4

Mistral AI

MM119.0B

Best score:0.7 (GPQA)

Released:Mar 2026

Price:$0.15/1M tokens

Pixtral-12B

Mistral AI

MM12.4B

Best score:0.7 (HumanEval)

Released:Sep 2024

Price:$0.15/1M tokens

Mistral Small 3.1 24B Base

Mistral AI

MM24.0B

Best score:0.8 (MMLU)

Released:Mar 2025

Price:$0.10/1M tokens

Mistral Small 3.2 24B Instruct

Mistral AI

MM23.6B

Best score:0.9 (HumanEval)

Released:Jun 2025

Mistral Small 3 24B Base

Mistral AI

MM23.6B

Best score:0.9 (ARC)

Released:Jan 2025

Mistral Small 3.1 24B Instruct

Mistral AI

MM24.0B

Best score:0.9 (HumanEval)

Released:Mar 2025

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.