vs
0:2
overall capability wins (0 ties across 2 scored). Mistral Small 4 leads overall.
- Pricing0:2
- Speed0:2
- Specs0:2
Verdict
Mistral Small 4 offers ~7.8x larger context (256K vs 33K); Mistral Small 4 allows ~7.8x more max output (64K vs 8K).
- Mistral Small 4 offers ~7.8x larger context (256K vs 33K)
- Mistral Small 4 allows ~7.8x more max output (64K vs 8K)
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Codestral 22B
- $0.35
- Mistral Small 4
- $0.30
Mistral Small 4 is about 1.2x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Mistral
Mistral
Family
Mistral
Mistral
License
MNPL
Apache 2.0
Open weights
Yes
Yes
Release
May 29, 2024
Mar 1, 2026
Knowledge cutoff
-
-
API / provider
Mistral / Together (ref.)
Mistral
Modalities
text → text
text, image → text
Specs
Context window
33K
256K
Max output
8K
64K
Parameters
22B
—
Pricing
Input $/1M
$0.20
$0.15
Output $/1M
$0.60
$0.60
Blended $/1M (3∶1)
$0.30
$0.26
Speed
tok/s
90
140
TTFT (s)
0.3
0.2
Reasoning
MMLU-Pro
—
—
GPQA Diamond
—
76.9
Humanity's Last Exam
—
9.9
AIME 2025
—
—
MATH-500
—
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
—
21.7
AA-LCR v1.1
—
49.7
CritPt
—
0.3
MMMU-Pro
—
56.8
IFBench
—
48.2
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
—
—
Terminal-Bench 2.1
—
21.0
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
41.8
—
SciCode
—
38.8
CursorBench
—
—
SWE-Rebench
—
—
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
—
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
—
—
| Metric | Codestral 22B | Mistral Small 4 | Delta |
|---|---|---|---|
| Identity | |||
| Organization | Mistral | Mistral | — |
| Family | Mistral | Mistral | — |
| License | MNPL | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | May 29, 2024 | Mar 1, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Mistral / Together (ref.) | Mistral | — |
| Modalities | text → text | text, image → text | — |
| Specs | |||
| Context window | 33K | 256K | B +223K |
| Max output | 8K | 64K | B +56K |
| Parameters | 22B | — | — |
| Pricing | |||
| Input $/1M | $0.20 | $0.15 | B +$0.05 |
| Output $/1M | $0.60 | $0.60 | tie |
| Blended $/1M (3∶1) | $0.30 | $0.26 | B +$0.04 |
| Speed | |||
| tok/s | 90 | 140 | B +50 |
| TTFT (s) | 0.3 | 0.2 | B +0.10 |
| Reasoning | |||
| MMLU-Pro | — | — | — |
| GPQA Diamond | — | 76.9 | — |
| Humanity's Last Exam | — | 9.9 | — |
| AIME 2025 | — | — | — |
| MATH-500 | — | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | — | 21.7 | — |
| AA-LCR v1.1 | — | 49.7 | — |
| CritPt | — | 0.3 | — |
| MMMU-Pro | — | 56.8 | — |
| IFBench | — | 48.2 | — |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | — | — | — |
| Terminal-Bench 2.1 | — | 21.0 | — |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | 41.8 | — | — |
| SciCode | — | 38.8 | — |
| CursorBench | — | — | — |
| SWE-Rebench | — | — | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | — | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | — | — | — |
Codestral 22B
- Context
- 33K / 8K out
- Parameters
- 22B
- Price
- $0.20 / $0.60
- Speed
- 90 tok/s · 0.3s TTFT
- Modalities
- text → text
- License
- MNPL
Original open-weight Codestral 22B — fill-in-the-middle coding model under Mistral's MNPL.
Mistral Small 4
- Context
- 256K / 64K out
- Parameters
- —
- Price
- $0.15 / $0.60
- Speed
- 140 tok/s · 0.2s TTFT
- Modalities
- text, image → text
- License
- Apache 2.0
Hybrid instruct/reasoning/coding Small model under Apache 2.0 - Mistral's high-throughput production default.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.
These two models have no overlapping published benchmarks in our dataset. Compare specs and pricing instead.