Mistral Medium 3.5
Mistral AI · released
A 128B open-weight model that folds chat, reasoning and coding into one set of weights. It is the default in Le Chat and the Vibe coding agent, and small enough to self-host on a handful of GPUs.
Best for: Self-hosting · Coding agents · European deployments
- Intelligence Index
- 14.2
- #35 of 38
- Price per 1M tokens
- $1.5 in $7.5 out
- Blended $3 at 3:1
- Context window
- 256K
- Output speed
- 168 tokens/s
- $0.50 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall20 · #35 of 38
Based on 2 of 3 benchmarks
- Coding5 · #38 of 39
Based on 3 of 5 benchmarks
- Agents & knowledge work13 · #35 of 40
Based on 2 of 6 benchmarks
- Reasoning & science4 · #36 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability29 · #35 of 39
Based on 2 of 2 benchmarks
- Vision17 · #30 of 31
Based on 2 of 2 benchmarks
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#35 of 38· Artificial Analysis
- Vals IndexNot tested
- Arena · Text1427 ±6
#28 of 33· 11,757 votes· Arena
Coding
#35 of 38· Artificial Analysis
- SciCode40.2%
#35 of 38· Artificial Analysis
- Arena · WebDev1263 ±15
#34 of 34· 2,325 votes· Arena
- Vibe Code BenchNot tested
- Code MigrationNot tested
Agents & knowledge work
- GDPval-AA13.2
#34 of 39· Artificial Analysis
- Arena · Agent-12.4 ±1.5
#29 of 29· Arena
- Finance Agent v2Not tested
- Excel Modeling BenchmarkNot tested
- Legal Research BenchNot tested
- Tax Agent BenchNot tested
Reasoning & science
- Humanity's Last Exam13.8%
#36 of 39· Artificial Analysis
- CritPt0.0%
#38 of 39· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA Diamondsaturating74.8%
#27 of 27· Artificial Analysis
Knowledge & reliability
#36 of 39· Artificial Analysis
- AA-LCR69.3%
#35 of 39· Artificial Analysis
Vision
- MMMU-Pro64.9%
#24 of 27· Artificial Analysis
- Arena · Vision1199 ±10
#19 of 19· 5,379 votes· Arena
Reported by Mistral AI
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- SWE-bench Verified77.6%
Specifications
- Lab
- Mistral AI
- Released
- Apr 28, 2026
- Availability
- Generally available
- Context window
- 256,000 tokens
- Max output
- —
- Accepts
- Text, Images
- Knowledge cutoff
- Not published
- Weights
- Open
- Licence
- Modified MIT
- Parameters
- 128B
- Reasoning control
- Effort none or high
- API model ID
mistral-medium-3-5
Pricing
- Input / 1M tokens
- $1.5
- Cached input / 1M
- $0.15
- Output / 1M tokens
- $7.5
- Blended (3:1)
- $3
- Measured cost / task
- $0.50
Batch processing is half price.
Worth knowing
- The Modified MIT licence carries restrictions for companies with large revenue.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.