Compare / head-to-head
Devstral 2vs
GPT-5.2
GPT-5.2 leads 2 of 2 shared benchmarks. Devstral 2 is 6.6x cheaper per token. GPT-5.2 has the larger context window (400K tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 2
GPT-5.2 leads
Cheaper per token
Devstral 2
6.6x cheaper, input + output
Larger context
GPT-5.2
400K tokens
Providers
2 providers
Mistral AI · OpenAI
Devstral 2
Mistral AI · released 2025-12-09
text
Quality
Benchmark matrix
2 shared · 2 only Devstral 2 · 34 only GPT-5.2| Benchmark | Devstral 2 | GPT-5.2 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 2 | ||||
| SWE-bench Verified | 72.2% ↗ | 80% ↗ | -7.8 pt | GPT-5.2 |
| LMArena WebDev (LMArena) | 1196.1 ↗ | 1415.5 ↗ | -219.4 | GPT-5.2 |
| Only Devstral 2 reports · 2 | ||||
| SWE-bench Multilingual | 61.3% ↗ | not reported | — | — |
| Terminal Bench 2 | 32.6% ↗ | not reported | — | — |
| Only GPT-5.2 reports · 34 | ||||
| GPQA Diamond | not reported | 92.4% ↗ | — | — |
| AIME 2025 | not reported | 100% ↗ | — | — |
| AIME 2026 | not reported | 98.3% ↗matharena | — | — |
| LMArena Elo | not reported | 1412.4 ↗ | — | — |
| ARC-AGI-1 (Verified) | not reported | 86.2% ↗ | — | — |
| ARC-AGI-2 (Verified) | not reported | 52.9% ↗ | — | — |
| Chess Puzzles (Epoch AI run) | not reported | 49% ↗epoch run | — | — |
| EBR-bench (Epoch AI run) | not reported | 23% ↗epoch run | — | — |
| FrontierMath Tier 4 v2 (Epoch AI run) | not reported | 31.7% ↗epoch run | — | — |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | not reported | 67.4% ↗epoch run | — | — |
| Furniture Assembly (Epoch AI run) | not reported | 38.3% ↗epoch run | — | — |
| GDPval (wins or ties) | not reported | 70.9% ↗ | — | — |
| GSO Opt@1 (GSO) | not reported | 26.5% ↗ | — | — |
| LiveBench Agentic Coding (LiveBench) | not reported | 50.3% ↗ | — | — |
| LiveBench Coding (LiveBench) | not reported | 76.1% ↗ | — | — |
| LiveBench Data Analysis (LiveBench) | not reported | 78.2% ↗ | — | — |
| LiveBench Instruction Following (LiveBench) | not reported | 61.8% ↗ | — | — |
| LiveBench Language (LiveBench) | not reported | 79.8% ↗ | — | — |
| LiveBench Mathematics (LiveBench) | not reported | 93.2% ↗ | — | — |
| LiveBench Reasoning (LiveBench) | not reported | 83.2% ↗ | — | — |
| LMArena Vision (LMArena) | not reported | 1242.8 ↗ | — | — |
| Mystery Game Puzzles (Epoch AI run) | not reported | 23% ↗epoch run | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | not reported | 96.1% ↗epoch run | — | — |
| SAGE (Vals AI) | not reported | 49.3% ↗ | — | — |
| SimpleBench (SimpleBench) | not reported | 45.8% ↗ | — | — |
| SimpleQA Verified | not reported | 37.1% ↗epoch run | — | — |
| SWE-Bench Pro | not reported | 55.6% ↗ | — | — |
| SWE-bench Verified (Epoch AI run) | not reported | 73.8% ↗epoch run | — | — |
| tau2-bench Airline (Sierra) | not reported | 83% ↗ | — | — |
| tau2-bench Banking Knowledge (Sierra) | not reported | 32.2% ↗ | — | — |
| tau2-bench Retail (Sierra) | not reported | 81.6% ↗ | — | — |
| tau2-bench Telecom (Sierra) | not reported | 89.7% ↗ | — | — |
| Vending-Bench 2 (Andon Labs) | not reported | 3591.33 ↗ | — | — |
| WeirdML (Håvard Tveit Ihle) | not reported | 72.2% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Devstral 2 minus GPT-5.2 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Devstral 2 | GPT-5.2 | Edge |
|---|---|---|---|
| Context window | 256K tokens | 400K tokens | GPT-5.2 |
| Max output | — | 128K tokens | — |
| Input price / 1M | $0.4 ↗ | $1.75 ↗ | Devstral 2 |
| Output price / 1M | $2 ↗ | $14 ↗ | Devstral 2 |
| Cached input / 1M | — | $0.175 ↗ | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $2.4 | $15.75 | Devstral 2 |
| Modalities | text | text · vision | GPT-5.2 |
| Released | 2025-12-09 | 2025-12-11 | — |
| Cited benchmark scores | 4 | 45 | — |
Reliability
Provider status
Live from /statusMore matchups
Devstral 2 vs …
Models sharing the most benchmarksMore matchups
GPT-5.2 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Devstral 2 and GPT-5.2 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.