vs
3:10
overall capability wins (2 ties across 15 scored). Mistral Large 3 leads overall.
- Reasoning1:6
- Coding1:3
- Tool use1:0
- Composite indices0:2
- Pricing3:0
- Speed2:0
- Specs0:1
Verdict
Mistral Large 3 leads coding (widest gap: +35.7 on WebDev Elo); Devstral 2 is ~3.8x cheaper on a blended token basis than Mistral Large 3.
- Mistral Large 3 leads coding (widest gap: +35.7 on WebDev Elo)
- Devstral 2 is ~3.8x cheaper on a blended token basis than Mistral Large 3
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Devstral 2
- $0.90
- Mistral Large 3
- $3.50
Devstral 2 is about 3.9x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Mistral
Mistral
Family
Mistral
Mistral
License
Apache 2.0
Apache 2.0
Open weights
Yes
Yes
Release
Dec 1, 2025
Nov 19, 2025
Knowledge cutoff
-
-
API / provider
Mistral
Mistral
Modalities
text → text
text, image → text
Specs
Context window
256K
256K
Max output
64K
64K
Parameters
123B
675B (41B act.)
Pricing
Input $/1M
$0.40
$2
Output $/1M
$2
$6
Blended $/1M (3∶1)
$0.80
$3
Speed
tok/s
85
70
TTFT (s)
0.35
0.45
Reasoning
MMLU-Pro
76.2
80.7
GPQA Diamond
59.4
68.0
Humanity's Last Exam
3.6
4.2
AIME 2025
36.7
38.0
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
20.8
25.0
AA-LCR v1.1
32.3
36.0
CritPt
—
—
MMMU-Pro
—
55.7
IFBench
38.1
36.2
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
44.8
46.5
Terminal-Bench 2.1
30.3
12.0
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
32.8
36.6
CursorBench
—
—
SWE-Rebench
42.0
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,194
1,230
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,413
| Metric | Devstral 2 | Mistral Large 3 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Mistral | Mistral | — |
| Family | Mistral | Mistral | — |
| License | Apache 2.0 | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Dec 1, 2025 | Nov 19, 2025 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Mistral | Mistral | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 256K | 256K | tie |
| Max output | 64K | 64K | tie |
| Parameters | 123B | 675B (41B act.) | B +552 |
| Pricing | |||
| Input $/1M | $0.40 | $2 | A +$1.60 |
| Output $/1M | $2 | $6 | A +$4 |
| Blended $/1M (3∶1) | $0.80 | $3 | A +$2.20 |
| Speed | |||
| tok/s | 85 | 70 | A +15 |
| TTFT (s) | 0.35 | 0.45 | A +0.10 |
| Reasoning | |||
| MMLU-Pro | 76.2 | 80.7 | B +4.5 pts |
| GPQA Diamond | 59.4 | 68.0 | B +8.6 pts |
| Humanity's Last Exam | 3.6 | 4.2 | B +0.6 pts |
| AIME 2025 | 36.7 | 38.0 | B +1.3 pts |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 20.8 | 25.0 | B +4.2 pts |
| AA-LCR v1.1 | 32.3 | 36.0 | B +3.7 pts |
| CritPt | — | — | — |
| MMMU-Pro | — | 55.7 | — |
| IFBench | 38.1 | 36.2 | A +1.9 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 44.8 | 46.5 | B +1.7 pts |
| Terminal-Bench 2.1 | 30.3 | 12.0 | A +18.3 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 32.8 | 36.6 | B +3.8 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 42.0 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,194 | 1,230 | B +36 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,413 | — |
Devstral 2
- Context
- 256K / 64K out
- Parameters
- 123B
- Price
- $0.40 / $2
- Speed
- 85 tok/s · 0.35s TTFT
- Modalities
- text → text
- License
- Apache 2.0
Open-weight agentic coding model for autonomous software engineering (successor path: Medium 3.5).
Mistral Large 3
- Context
- 256K / 64K out
- Parameters
- 675B (41B act.)
- Price
- $2 / $6
- Speed
- 70 tok/s · 0.45s TTFT
- Modalities
- text, image → text
- License
- Apache 2.0
Mistral's open Large 3 MoE - multilingual European frontier alternative under Apache 2.0.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.