Codestral 2508 vs GPT-5.5
Updated October 21, 2026. 7 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
vs boards won of 7 shared
Decision reading
Only GPT-5.5 has a coding score, 71.3 on the Vibe Coding Index; Codestral 2508 has none, so the two cannot be ranked on coding.
Codestral 2508 has the edge where output tokens cost 33.3x less.
Model ACodestral 2508Mistral AINot benchmarkedNo coding benchmarkNo cost per solved taskModel BGPT-5.5OpenAI71.3/1005 independent benchmarks$1.65 per solved task
- Shared results
- 7
- Codestral 2508 only
- 0
- GPT-5.5 only
- 125
- Like-for-like categories
- 2 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only GPT-5.5 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
No call
Only GPT-5.5 has a AA Intelligence Index result, 38.4.
Directional only
Agentic work
AA Agentic Index
No call
Only GPT-5.5 has a AA Agentic Index result, 36.4.
Directional only
Reasoning
GPQA Diamond
No call
Only GPT-5.5 has a GPQA Diamond result, 93.2%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5.5 has a MMLU Pro result, 88.1%.
Directional only
Web design
Design Arena: websites
GPT-5.5
GPT-5.5 leads, 1256 against 1013.
Same board, both measured
Long documents
Listed context window
GPT-5.5
GPT-5.5 takes 1.05M tokens against 256K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Codestral 2508
Codestral 2508 $0.0008, GPT-5.5 $0.020 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5.5 has the number, $1.65.
Directional only
At a glance
- Coding index
- Codestral 2508 not benchmarked, GPT-5.5 71.3 from 5 benchmarks.
- Widest category
- Coding, 60.2 points apart on average over 1 shared benchmark.
- Widest single gap
- ALE-Bench: Codestral 2508 137.8 against GPT-5.5 1943.0, 60.2 points apart on a 0 to 100 scale.
- Output price
- Codestral 2508 $0.90 against GPT-5.5 $30.00 per million output tokens, 33.3x apart.
- Context window
- Codestral 2508 takes 256K tokens, GPT-5.5 1.05M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Codestral 2508 | Bars | GPT-5.5 | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Web design | 0 | 6 | ||
| All shared boards | 0 | 7 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index01
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 85.1% | Only GPT-5.5 measured |
| LMArena WebDev | Not measured | 1513 | Only GPT-5.5 measured |
| AA Coding Index | Not measured | 74.9 | Only GPT-5.5 measured |
| LiveCodeBench (Vals AI) | Not measured | 85.3% | Only GPT-5.5 measured |
| SWE-bench Verified (Vals AI) | Not measured | 82.6% | Only GPT-5.5 measured |
| Vibe Code Bench (Vals AI) | Not measured | 69.8% | Only GPT-5.5 measured |
| Code migration (Vals AI) | Not measured | 45.2% | Only GPT-5.5 measured |
| ProgramBench (Vals AI) | Not measured | 0.5% | Only GPT-5.5 measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 76.4% | Only GPT-5.5 measured |
| SkillsBench (Vals AI) | Not measured | 62.5% | Only GPT-5.5 measured |
| SciCode | Not measured | 56.1% | Only GPT-5.5 measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 60.6% | Only GPT-5.5 measured |
| Terminal-Bench | Not measured | 84.7% | Only GPT-5.5 measured |
| FrontierCode | Not measured | 43.0% | Only GPT-5.5 measured |
| GSO-Bench | Not measured | 40.2% | Only GPT-5.5 measured |
| WeirdML | Not measured | 84.9% | Only GPT-5.5 measured |
| MirrorCode | Not measured | 10.0% | Only GPT-5.5 measured |
| ALE-Bench | 137.8 | 1943.0 | GPT-5.5 |
| LMArena Text: coding | Not measured | 1494 | Only GPT-5.5 measured |
| LMArena Text: software and IT | Not measured | 1493 | Only GPT-5.5 measured |
| LMArena WebDev: React | Not measured | 1506 | Only GPT-5.5 measured |
| LMArena WebDev: HTML | Not measured | 1543 | Only GPT-5.5 measured |
| LMArena WebDev: from an image | Not measured | 1532 | Only GPT-5.5 measured |
| Neither model has a result on the other 15 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 38.4 | Only GPT-5.5 measured |
| Epoch Capabilities Index | Not measured | 159.7 | Only GPT-5.5 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index25 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 36.4 | Only GPT-5.5 measured |
| Design Arena: agents building web apps | Not measured | 1117 | Only GPT-5.5 measured |
| Design Arena: agents building full-stack apps | Not measured | 1089 | Only GPT-5.5 measured |
| Design Arena: agents building mobile apps | Not measured | 1142 | Only GPT-5.5 measured |
| Design Arena: agents building Android apps | Not measured | 1169 | Only GPT-5.5 measured |
| Design Arena: agents building Godot games | Not measured | 1173 | Only GPT-5.5 measured |
| Design Arena: agents building games | Not measured | 1179 | Only GPT-5.5 measured |
| Design Arena: HTML slides | Not measured | 1067 | Only GPT-5.5 measured |
| Design Arena: PowerPoint slides | Not measured | 1109 | Only GPT-5.5 measured |
| LiveBench research | Not measured | 61.6% | Only GPT-5.5 measured |
| LiveBench browser | Not measured | 48.0% | Only GPT-5.5 measured |
| LiveBench long agentic | Not measured | 44.6% | Only GPT-5.5 measured |
| SREBench (Vals AI) | Not measured | 3.8% | Only GPT-5.5 measured |
| Finance Agent (Vals AI) | Not measured | 51.8% | Only GPT-5.5 measured |
| Tax Agent Bench (Vals AI) | Not measured | 21.8% | Only GPT-5.5 measured |
| Legal Research Bench (Vals AI) | Not measured | 40.4% | Only GPT-5.5 measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 64.5% | Only GPT-5.5 measured |
| Time Horizon Index (Vals AI) | Not measured | 8.3% | Only GPT-5.5 measured |
| τ²-Bench Telecom | Not measured | 93.9% | Only GPT-5.5 measured |
| APEX-Agents | Not measured | 55.1% | Only GPT-5.5 measured |
| Remote Labor Index | Not measured | 6.3% | Only GPT-5.5 measured |
| OSWorld 2.0 | Not measured | 13.0% | Only GPT-5.5 measured |
| DeepResearch Bench | Not measured | 54.0% | Only GPT-5.5 measured |
| PostTrainBench | Not measured | 27.2% | Only GPT-5.5 measured |
| Vending-Bench 2 | Not measured | $7,524 | Only GPT-5.5 measured |
| Neither model has a result on the other 14 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index17 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 93.2% | Only GPT-5.5 measured |
| LiveBench reasoning | Not measured | 89.7% | Only GPT-5.5 measured |
| MMMU Pro (Vals AI) | Not measured | 88.3% | Only GPT-5.5 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 93.5% | Only GPT-5.5 measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 45.8% | Only GPT-5.5 measured |
| GPQA Diamond (Epoch AI run) | Not measured | 90.7% | Only GPT-5.5 measured |
| ARC-AGI | Not measured | 95.0% | Only GPT-5.5 measured |
| ARC-AGI-2 | Not measured | 85.0% | Only GPT-5.5 measured |
| SimpleBench | Not measured | 69.0% | Only GPT-5.5 measured |
| Chess puzzles | Not measured | 63.0% | Only GPT-5.5 measured |
| Mystery game puzzles | Not measured | 56.0% | Only GPT-5.5 measured |
| EBR-bench | Not measured | 34.3% | Only GPT-5.5 measured |
| LMArena Text: hard prompts | Not measured | 1489 | Only GPT-5.5 measured |
| LMArena Text: expert | Not measured | 1507 | Only GPT-5.5 measured |
| LMArena Text: hard prompts in English | Not measured | 1483 | Only GPT-5.5 measured |
| LMArena Vision: diagrams | Not measured | 1325 | Only GPT-5.5 measured |
| LMArena Vision: homework | Not measured | 1332 | Only GPT-5.5 measured |
| Neither model has a result on the other 4 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| OTIS Mock AIME | Not measured | 84.4% | Only GPT-5.5 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 85.3% | Only GPT-5.5 measured |
| FrontierMath, tier 4 | Not measured | 72.5% | Only GPT-5.5 measured |
| ProofBench | Not measured | 50.0% | Only GPT-5.5 measured |
| LMArena Text: math | Not measured | 1487 | Only GPT-5.5 measured |
| LMArena Text: mathematical work | Not measured | 1496 | Only GPT-5.5 measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index12 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 88.1% | Only GPT-5.5 measured |
| LegalBench (Vals AI) | Not measured | 86.5% | Only GPT-5.5 measured |
| CaseLaw (Vals AI) | Not measured | 66.2% | Only GPT-5.5 measured |
| CorpFin (Vals AI) | Not measured | 68.4% | Only GPT-5.5 measured |
| TaxEval (Vals AI) | Not measured | 75.0% | Only GPT-5.5 measured |
| MortgageTax (Vals AI) | Not measured | 68.8% | Only GPT-5.5 measured |
| MedCode (Vals AI) | Not measured | 49.1% | Only GPT-5.5 measured |
| SimpleQA Verified | Not measured | 63.0% | Only GPT-5.5 measured |
| LMArena Text: science | Not measured | 1485 | Only GPT-5.5 measured |
| LMArena Text: medicine and healthcare | Not measured | 1465 | Only GPT-5.5 measured |
| LMArena Text: legal and government | Not measured | 1491 | Only GPT-5.5 measured |
| LMArena Text: business and finance | Not measured | 1478 | Only GPT-5.5 measured |
| Neither model has a result on the other 5 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index06
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| Design Arena: websites | 1013 | 1256 | GPT-5.5 |
| Design Arena: UI components | 1017 | 1260 | GPT-5.5 |
| Design Arena: game dev | 968 | 1294 | GPT-5.5 |
| Design Arena: data viz | 1017 | 1264 | GPT-5.5 |
| Design Arena: 3D | 1020 | 1197 | GPT-5.5 |
| Design Arena: all code categories | 1007 | 1260 | GPT-5.5 |
| Design Arena: SVG | Not measured | 1234 | Only GPT-5.5 measured |
| Design Arena: ASCII art | Not measured | 1255 | Only GPT-5.5 measured |
| LiveBench design | Not measured | 70.3% | Only GPT-5.5 measured |
GeneralDirectional onlyPublic boards, shown as published, kept out of the index38 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| LiveBench data analysis | Not measured | 81.6% | Only GPT-5.5 measured |
| Public Benefits Bench (Vals AI) | Not measured | 60.9% | Only GPT-5.5 measured |
| SAGE (Vals AI) | Not measured | 51.5% | Only GPT-5.5 measured |
| MedScribe (Vals AI) | Not measured | 86.9% | Only GPT-5.5 measured |
| Vals Multimodal Index | Not measured | 68.1% | Only GPT-5.5 measured |
| IFBench | Not measured | 75.9% | Only GPT-5.5 measured |
| Long context reasoning (AA-LCR) | Not measured | 84.3% | Only GPT-5.5 measured |
| DTBench | Not measured | 96.0% | Only GPT-5.5 measured |
| LMCA | Not measured | 54.3% | Only GPT-5.5 measured |
| Surface Evolver Bench | Not measured | 88.1% | Only GPT-5.5 measured |
| Furniture assembly | Not measured | 44.2% | Only GPT-5.5 measured |
| ExploitBench | Not measured | 47.4% | Only GPT-5.5 measured |
| CL-bench Life | Not measured | 22.2% | Only GPT-5.5 measured |
| LMArena Text | Not measured | 1471 | Only GPT-5.5 measured |
| LMArena Text: instruction following | Not measured | 1478 | Only GPT-5.5 measured |
| LMArena Text: creative writing | Not measured | 1455 | Only GPT-5.5 measured |
| LMArena Vision | Not measured | 1297 | Only GPT-5.5 measured |
| LMArena Text: English | Not measured | 1470 | Only GPT-5.5 measured |
| LMArena Text: non-English | Not measured | 1467 | Only GPT-5.5 measured |
| LMArena Text: Chinese | Not measured | 1534 | Only GPT-5.5 measured |
| LMArena Text: German | Not measured | 1480 | Only GPT-5.5 measured |
| LMArena Text: Japanese | Not measured | 1499 | Only GPT-5.5 measured |
| LMArena Text: Korean | Not measured | 1460 | Only GPT-5.5 measured |
| LMArena Text: French | Not measured | 1485 | Only GPT-5.5 measured |
| LMArena Text: Spanish | Not measured | 1468 | Only GPT-5.5 measured |
| LMArena Text: Russian | Not measured | 1473 | Only GPT-5.5 measured |
| LMArena Text: Polish | Not measured | 1465 | Only GPT-5.5 measured |
| LMArena Text: multi-turn | Not measured | 1476 | Only GPT-5.5 measured |
| LMArena Text: longer queries | Not measured | 1484 | Only GPT-5.5 measured |
| LMArena Text: writing and language | Not measured | 1466 | Only GPT-5.5 measured |
| LMArena Text: entertainment and media | Not measured | 1449 | Only GPT-5.5 measured |
| LMArena Vision: English | Not measured | 1292 | Only GPT-5.5 measured |
| LMArena Vision: Chinese | Not measured | 1360 | Only GPT-5.5 measured |
| LMArena Vision: reading text | Not measured | 1306 | Only GPT-5.5 measured |
| LMArena Vision: humor | Not measured | 1293 | Only GPT-5.5 measured |
| LMArena Vision: recognizing things | Not measured | 1231 | Only GPT-5.5 measured |
| LMArena Vision: creative writing | Not measured | 1301 | Only GPT-5.5 measured |
| LMArena Document | Not measured | 1486 | Only GPT-5.5 measured |
| Neither model has a result on the other 8 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Codestral 2508 | GPT-5.5 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Codestral 2508$0.0008Fits in one request
GPT-5.5$0.020Fits in one request
Codestral 2508 costs less
Repository review
50K input + 3K output tokens
Codestral 2508$0.018Fits in one request
GPT-5.5$0.340Fits in one request
Codestral 2508 costs less
Agent loop
200K input + 10K output tokens
Codestral 2508$0.069Fits in one request
GPT-5.5$1.30Fits in one request
Codestral 2508 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Mistral Codestral 2508 | OpenAI GPT-5.5 |
|---|---|---|
| Maker | Mistral AI | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | August 1, 2025 | April 24, 2026 |
| Context window | 256K | 1.05M |
| Input $/Mtok | $0.30 | $5.00 |
| Output $/Mtok | $0.90 | $30.00 |
| Coding index | Not benchmarked | 71.3 |
| Independent measurements behind it | 0 | 5 |
| Cost per solved task | Not computable | $1.65 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Codestral 2508 better than GPT-5.5 for coding?
Only GPT-5.5 has a coding score, 71.3 on the Vibe Coding Index; Codestral 2508 has none, so the two cannot be ranked on coding. Codestral 2508 has the edge where output tokens cost 33.3x less.
Which is cheaper, Codestral 2508 or GPT-5.5?
Codestral 2508 costs $0.30 per million input tokens and $0.90 per million output tokens; GPT-5.5 costs $5.00 and $30.00. Cost per solved task is known for GPT-5.5 only.
Which has the bigger context window, Codestral 2508 or GPT-5.5?
GPT-5.5, with 1.05M tokens against 256K.
How many benchmarks is the Codestral 2508 vs GPT-5.5 comparison based on?
7 benchmarks have a result for both models as of October 21, 2026. Codestral 2508's coding index rests on 0 independent benchmarks and GPT-5.5's on 5. Where only one model has a result, the page says so instead of naming a winner.