vs
0:4
overall capability wins (0 ties across 4 scored). Mistral Medium 3.5 leads overall.
- Reasoning0:1
- Pricing3:0
- Speed1:1
- Specs0:3
Verdict
Mistral Medium 3.5 edges reasoning & knowledge; Mistral Medium 3.5 offers ~2.0x larger context (256K vs 128K); Mistral Medium 3.5 allows ~1.6x more max output (64K vs 40K).
- Mistral Medium 3.5 edges reasoning & knowledge
- Mistral Medium 3.5 offers ~2.0x larger context (256K vs 128K)
- Mistral Medium 3.5 allows ~1.6x more max output (64K vs 40K)
- Magistral Small is ~4.0x 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.
- Magistral Small
- $0.88
- Mistral Medium 3.5
- $3.38
Magistral Small is about 3.9x 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
Jun 10, 2025
Apr 22, 2026
Knowledge cutoff
-
-
API / provider
Mistral
Mistral
Modalities
text, image → text
text, image → text
Specs
Context window
128K
256K
Max output
40K
64K
Parameters
24B
128B
Pricing
Input $/1M
$0.50
$1.50
Output $/1M
$1.50
$7.50
Blended $/1M (3∶1)
$0.75
$3
Speed
tok/s
90
80
TTFT (s)
0.55
0.4
Reasoning
MMLU-Pro
—
—
GPQA Diamond
68.2
74.8
Humanity's Last Exam
—
13.8
AIME 2025
62.8
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
24.7
AA-LCR v1.1
—
69.3
CritPt
—
—
MMMU-Pro
—
64.9
IFBench
—
68.8
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
55.8
—
Terminal-Bench 2.1
—
50.6
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
—
40.2
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,265
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
1,426
| Metric | Magistral Small | Mistral Medium 3.5 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Mistral | Mistral | — |
| Family | Mistral | Mistral | — |
| License | Apache 2.0 | Modified MIT | — |
| Open weights | Yes | Yes | — |
| Release | Jun 10, 2025 | Apr 22, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Mistral | Mistral | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 256K | B +128K |
| Max output | 40K | 64K | B +24K |
| Parameters | 24B | 128B | B +104 |
| Pricing | |||
| Input $/1M | $0.50 | $1.50 | A +$1 |
| Output $/1M | $1.50 | $7.50 | A +$6 |
| Blended $/1M (3∶1) | $0.75 | $3 | A +$2.25 |
| Speed | |||
| tok/s | 90 | 80 | A +10 |
| TTFT (s) | 0.55 | 0.4 | B +0.15 |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | 68.2 | 74.8 | B +6.6 pts |
| Humanity's Last Exam | — | 13.8 | — |
| AIME 2025 | 62.8 | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 24.7 | — |
| AA-LCR v1.1 | — | 69.3 | — |
| CritPt | — | — | — |
| MMMU-Pro | — | 64.9 | — |
| IFBench | — | 68.8 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 55.8 | — | — |
| Terminal-Bench 2.1 | — | 50.6 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | — | 40.2 | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,265 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | 1,426 | — |
Magistral Small
- Context
- 128K / 40K out
- Parameters
- 24B
- Price
- $0.50 / $1.50
- Speed
- 90 tok/s · 0.55s TTFT
- Modalities
- text, image → text
- License
- Apache 2.0
Open-weight 24B Magistral reasoning model (Apache 2.0); deprecated on Mistral API in favor of Small 4 / Medium 3.5 reasoning modes.
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.