vs
0:10
overall capability wins (2 ties across 12 scored). Mistral Medium 3.5 leads overall.
- Reasoning0:5
- Coding0:3
- Tool use0:1
- Composite indices0:2
- Pricing3:0
- Speed2:0
- Specs0:1
Verdict
Mistral Medium 3.5 leads coding (widest gap: +71.1 on WebDev Elo); Devstral 2 is ~3.8x cheaper on a blended token basis than Mistral Medium 3.5.
- Mistral Medium 3.5 leads coding (widest gap: +71.1 on WebDev Elo)
- Devstral 2 is ~3.8x cheaper on a blended token basis than Mistral Medium 3.5
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Devstral 2
- $0.90
- Mistral Medium 3.5
- $3.38
Devstral 2 is about 3.8x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Mistral
Mistral
Family
Mistral
Mistral
License
Apache 2.0
Modified MIT
Open weights
Yes
Yes
Release
Dec 1, 2025
Apr 22, 2026
Knowledge cutoff
-
-
API / provider
Mistral
Mistral
Modalities
text → text
text, image → text
Specs
Context window
256K
256K
Max output
64K
64K
Parameters
123B
128B
Pricing
Input $/1M
$0.40
$1.50
Output $/1M
$2
$7.50
Blended $/1M (3∶1)
$0.80
$3
Speed
tok/s
85
80
TTFT (s)
0.35
0.4
Reasoning
MMLU-Pro
76.2
—
GPQA Diamond
59.4
74.8
Humanity's Last Exam
3.6
13.8
AIME 2025
36.7
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
20.8
24.7
AA-LCR v1.1
32.3
69.3
CritPt
—
—
MMMU-Pro
—
64.9
IFBench
38.1
68.8
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
44.8
—
Terminal-Bench 2.1
30.3
50.6
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
32.8
40.2
CursorBench
—
—
SWE-Rebench
42.0
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
1,194
1,265
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,426
| Metric | Devstral 2 | Mistral Medium 3.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Mistral | Mistral | — |
| Family | Mistral | Mistral | — |
| License | Apache 2.0 | Modified MIT | — |
| Open weights | Yes | Yes | — |
| Release | Dec 1, 2025 | Apr 22, 2026 | — |
| 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 | 128B | B +5.00 |
| Pricing | |||
| Input $/1M | $0.40 | $1.50 | A +$1.10 |
| Output $/1M | $2 | $7.50 | A +$5.50 |
| Blended $/1M (3∶1) | $0.80 | $3 | A +$2.20 |
| Speed | |||
| tok/s | 85 | 80 | A +5.00 |
| TTFT (s) | 0.35 | 0.4 | A +0.05 |
| Reasoning | |||
| MMLU-Pro | 76.2 | — | — |
| GPQA Diamond | 59.4 | 74.8 | B +15.4 pts |
| Humanity's Last Exam | 3.6 | 13.8 | B +10.2 pts |
| AIME 2025 | 36.7 | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 20.8 | 24.7 | B +3.9 pts |
| AA-LCR v1.1 | 32.3 | 69.3 | B +37.0 pts |
| CritPt | — | — | — |
| MMMU-Pro | — | 64.9 | — |
| IFBench | 38.1 | 68.8 | B +30.7 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 44.8 | — | — |
| Terminal-Bench 2.1 | 30.3 | 50.6 | B +20.3 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 32.8 | 40.2 | B +7.4 pts |
| CursorBench | — | — | — |
| SWE-Rebench | 42.0 | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | 1,194 | 1,265 | B +71 Elo |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,426 | — |
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 Medium 3.5
- Context
- 256K / 64K out
- Parameters
- 128B
- Price
- $1.50 / $7.50
- Speed
- 80 tok/s · 0.4s TTFT
- Modalities
- text, image → text
- License
- Modified MIT
Mistral's April 2026 frontier multimodal model (open weights, Modified MIT) powering Vibe remote agents at $1.50/$7.50.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.