Mistral Large 4
Mistral AI · released
Mistral's new trillion-parameter multimodal flagship, currently an API preview. It targets coding agents, cybersecurity, finance and legal work and document understanding, with weights promised so customers can host it themselves.
Best for: Agentic workflows · Document and image understanding · European data residency
- Intelligence Index
- 38.4
- #20 of 38
- Price per 1M tokens
- $1.36 in $4.18 out
- Blended $2.06 at 3:1
- Context window
- 1M
- Output speed
- 116 tokens/s
- $1.13 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.
- Overall56 · #23 of 38
Based on 2 of 3 benchmarks
- Coding49 · #22 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work57 · #23 of 40
Based on 5 of 6 benchmarks
- Reasoning & science41 · #24 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability60 · #26 of 39
Based on 2 of 2 benchmarks
- Vision63 · #18 of 31
Based on 1 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
#20 of 38· Artificial Analysis
- Vals Index48.1%
#21 of 26· High effort· $13.8 per test· Vals AI
- Arena · TextNot tested
Coding
- Terminal-Bench 4.026.8%
#15 of 38· Artificial Analysis
- SciCode54.2%
#17 of 38· Artificial Analysis
- Arena · WebDev1534 ±21
#23 of 34· 872 votes· Arena
- Vibe Code Bench78.4%
#17 of 26· High effort· Vals AI
- Code Migration22.8%
#21 of 26· High effort· Vals AI
Agents & knowledge work
- GDPval-AA46.2
#23 of 39· Artificial Analysis
- Arena · AgentNot tested
- Finance Agent v254.7%
#11 of 26· High effort· Vals AI
#21 of 26· High effort· Vals AI
- Legal Research Bench31.7%
#21 of 26· High effort· Vals AI
- Tax Agent Bench63.3%
#15 of 26· High effort· Vals AI
Reasoning & science
- Humanity's Last Exam35.0%
#25 of 39· Artificial Analysis
- CritPt10.6%
#24 of 39· Artificial Analysis
- ARC-AGI-2saturatingNot tested
- ARC-AGI-3Not tested
- GPQA DiamondsaturatingNot tested
Knowledge & reliability
#26 of 39· Artificial Analysis
- AA-LCR81.3%
#17 of 39· Artificial Analysis
Vision
- MMMU-Pro76.4%
#16 of 27· Artificial Analysis
- Arena · VisionNot tested
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.
- Terminal-Bench 4.0 · preview28.3%
Specifications
- Lab
- Mistral AI
- Released
- Oct 6, 2026
- Availability
- Preview
- Context window
- 1,000,000 tokens
- Max output
- —
- Accepts
- Text, Images
- Knowledge cutoff
- Not published
- Weights
- Closed
- Parameters
- 1.05T total · 52B active
- Reasoning control
- Effort none or high
- API model ID
mistral-large-4-0
Public API preview. Mistral has said it will release the weights by the end of October; until they are published the model is listed here as closed.
Pricing
- Input / 1M tokens
- $1.36
- Cached input / 1M
- $0.14
- Output / 1M tokens
- $4.18
- Blended (3:1)
- $2.06
- Measured cost / task
- $1.13
List price. Mistral is currently running a 50% launch discount ($0.68 in / $2.09 out) with no stated end date.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.