Atlas

Models

← All models

Mistral Medium 3.5

Mistral AI · 2026-04-29 · 127.7B parameters

Mistral Medium 3.5 is Mistral AI's dense, 128B-class open-weight multimodal model, released in April 2026 and optimized for agentic and coding use cases. It accepts text and image input, produces text, supports a 256K-token context window, and offers per-request configurable reasoning effort alongside native function calling and JSON output. The weights use Mistral's Modified MIT license, and Mistral says the model can be self-hosted on as few as four GPUs.

Benchmark scores

BenchmarkScore
AA-Briefcase Elo516
AA-LCR61.0
AA-Omniscience Index-36.3
ALE-Bench1131
Artificial Analysis Agentic Index19.0
Artificial Analysis Coding Index46.9
Artificial Analysis Intelligence Index29.9
Artificial Analysis Omniscience Accuracy25.1
Artificial Analysis Omniscience Hallucination Rate82.0
Artificial Analysis Openness Index33.3
AutomationBench-AA13.7
BLXBench70.3
BLXBench Coding category88.9
BLXBench Cost category90.0
BLXBench Debugging category69.6
BLXBench Hallucination category70.6
BLXBench pass rate48.1
BLXBench Reasoning category67.8
BLXBench Refactoring category61.2
BLXBench Security category60.2
BLXBench Speed category92.2
BLXBench UI category76.9
BuseyBench SVG2.4
Buyout Game Bradley-Terry Rating1332
Capability142.7
CaseLaw44.2
CorpFin v258.8
CorpFin v2 - Exact Pages55.4
CorpFin v2 - Max Fitting Context61.1
CorpFin v2 - Shared Max Context59.9
Loading Atlas data…