Claude Opus 4.8 vs GPT-5.2
Updated October 21, 2026. 93 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Claude Opus 4.8 for coding: it scores 66.0 on the Vibe Coding Index against 58.9 for GPT-5.2.
Pick GPT-5.2 when output tokens cost 1.8x less, or when a solved task costs $1.91 against $2.17.
- Shared results
- 93
- Claude Opus 4.8 only
- 33
- GPT-5.2 only
- 18
- Like-for-like categories
- 8 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, four public coding benchmarks
Claude Opus 4.8
Claude Opus 4.8 scores 66.0 against 58.9.
2 benchmarks behind each score
Intelligence
AA Intelligence Index
Claude Opus 4.8
Claude Opus 4.8 leads, 41.8 against 30.4.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Claude Opus 4.8 has a AA Agentic Index result, 41.9.
Directional only
Reasoning
GPQA Diamond
Claude Opus 4.8
Claude Opus 4.8 leads, 92.4% against 91.7%.
Same board, both measured
Math
AIME
No call
Only GPT-5.2 has a AIME result, 96.9%.
Directional only
Knowledge
MMLU Pro
Claude Opus 4.8
Claude Opus 4.8 leads, 89.6% against 86.2%.
Same board, both measured
Web design
Design Arena: websites
Claude Opus 4.8
Claude Opus 4.8 leads, 1258 against 1196.
Same board, both measured
Long documents
Listed context window
Claude Opus 4.8
Claude Opus 4.8 takes 1M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.2
Claude Opus 4.8 $0.018, GPT-5.2 $0.0088 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.2
Claude Opus 4.8 $2.17, GPT-5.2 $1.91.
Published attempt cost and pass rate
At a glance
- Coding index
- Claude Opus 4.8 66.0 from 2 benchmarks, GPT-5.2 58.9 from 2 benchmarks.
- Widest category
- Math, 18.4 points apart on average over 6 shared benchmarks.
- Widest single gap
- ProofBench: Claude Opus 4.8 69.0% against GPT-5.2 15.0%, 54.0 points apart on a 0 to 100 scale.
- Output price
- Claude Opus 4.8 $25.00 against GPT-5.2 $14.00 per million output tokens, 1.8x apart.
- Context window
- Claude Opus 4.8 takes 1M tokens, GPT-5.2 400K.
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 | Claude Opus 4.8 | Bars | GPT-5.2 | Ties |
|---|---|---|---|---|
| Coding | 10 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 9 | 0 | ||
| Reasoning | 14 | 2 | ||
| Math | 6 | 0 | ||
| Knowledge | 10 | 1 | ||
| Web design | 8 | 0 | ||
| General | 29 | 2 | ||
| All shared boards | 88 | 5 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index100
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| SWE-bench Verified | Not measured | 72.8% | Only GPT-5.2 measured |
| SWE-bench Multilingual | Not measured | 66.7% | Only GPT-5.2 measured |
| LiveBench coding | 77.3% | Not measured | Only Claude Opus 4.8 measured |
| LMArena WebDev | 1556 | 1417 | Claude Opus 4.8 |
| AA Coding Index | 74.3 | Not measured | Only Claude Opus 4.8 measured |
| LiveCodeBench (Vals AI) | 87.8% | 85.4% | Claude Opus 4.8 |
| SWE-bench Verified (Vals AI) | 88.6% | 75.8% | Claude Opus 4.8 |
| Terminal-Bench 4 (Vals AI) | 23.2% | Not measured | Only Claude Opus 4.8 measured |
| Vibe Code Bench (Vals AI) | 82.7% | 53.5% | Claude Opus 4.8 |
| Code migration (Vals AI) | 47.3% | Not measured | Only Claude Opus 4.8 measured |
| ProgramBench (Vals AI) | 1.0% | Not measured | Only Claude Opus 4.8 measured |
| Terminal-Bench 2.1 (Vals AI) | 71.9% | Not measured | Only Claude Opus 4.8 measured |
| SkillsBench (Vals AI) | 59.2% | Not measured | Only Claude Opus 4.8 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 89.4% | Only GPT-5.2 measured |
| SciCode | 54.4% | Not measured | Only Claude Opus 4.8 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 58.3% | 47.0% | Claude Opus 4.8 |
| Terminal-Bench | Not measured | 64.9% | Only GPT-5.2 measured |
| FrontierCode | 46.5% | Not measured | Only Claude Opus 4.8 measured |
| GSO-Bench | 47.1% | 27.5% | Claude Opus 4.8 |
| WeirdML | 82.9% | 72.2% | Claude Opus 4.8 |
| ALE-Bench | Not measured | 1293.5 | Only GPT-5.2 measured |
| LMArena Text: coding | 1490 | 1442 | Claude Opus 4.8 |
| LMArena Text: software and IT | 1488 | 1434 | Claude Opus 4.8 |
| LMArena WebDev: React | 1557 | Not measured | Only Claude Opus 4.8 measured |
| LMArena WebDev: HTML | 1543 | 1427 | Claude Opus 4.8 |
| Neither model has a result on the other 13 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| AA Intelligence Index | 41.8 | 30.4 | Claude Opus 4.8 |
| Epoch Capabilities Index | 159.0 | 153.7 | Claude Opus 4.8 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index90
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| AA Agentic Index | 41.9 | Not measured | Only Claude Opus 4.8 measured |
| Design Arena: agents building web apps | 1233 | 1086 | Claude Opus 4.8 |
| Design Arena: agents building full-stack apps | 1249 | 1043 | Claude Opus 4.8 |
| Design Arena: agents building mobile apps | 1222 | 1094 | Claude Opus 4.8 |
| Design Arena: agents building Android apps | 1237 | 1064 | Claude Opus 4.8 |
| Design Arena: agents building Godot games | 1231 | 1142 | Claude Opus 4.8 |
| Design Arena: agents building games | 1238 | Not measured | Only Claude Opus 4.8 measured |
| Design Arena: HTML slides | 1221 | Not measured | Only Claude Opus 4.8 measured |
| Design Arena: PowerPoint slides | 1227 | Not measured | Only Claude Opus 4.8 measured |
| LiveBench research | 81.8% | Not measured | Only Claude Opus 4.8 measured |
| LiveBench browser | 96.0% | Not measured | Only Claude Opus 4.8 measured |
| LiveBench long agentic | 48.3% | Not measured | Only Claude Opus 4.8 measured |
| Finance Agent (Vals AI) | 53.9% | Not measured | Only Claude Opus 4.8 measured |
| Tax Agent Bench (Vals AI) | 30.0% | Not measured | Only Claude Opus 4.8 measured |
| Legal Research Bench (Vals AI) | 43.8% | Not measured | Only Claude Opus 4.8 measured |
| Excel Modeling Benchmark (Vals AI) | 69.4% | Not measured | Only Claude Opus 4.8 measured |
| Terminal-Bench-Science (Vals AI) | 4.3% | Not measured | Only Claude Opus 4.8 measured |
| Time Horizon Index (Vals AI) | 11.8% | Not measured | Only Claude Opus 4.8 measured |
| τ²-Bench Telecom | 94.4% | 84.8% | Claude Opus 4.8 |
| APEX-Agents | 48.9% | Not measured | Only Claude Opus 4.8 measured |
| METR time horizons | Not measured | 75.3% | Only GPT-5.2 measured |
| Remote Labor Index | 8.3% | 2.5% | Claude Opus 4.8 |
| GDPval | Not measured | 49.7% | Only GPT-5.2 measured |
| OSWorld 2.0 | 20.6% | Not measured | Only Claude Opus 4.8 measured |
| DeepResearch Bench | 50.2% | 41.1% | Claude Opus 4.8 |
| PostTrainBench | 33.8% | Not measured | Only Claude Opus 4.8 measured |
| Vending-Bench 2 | $5,787 | $3,591 | Claude Opus 4.8 |
| LMArena Search | 1204 | Not measured | Only Claude Opus 4.8 measured |
| Neither model has a result on the other 11 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index142
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| GPQA Diamond | 92.4% | 91.7% | Claude Opus 4.8 |
| LiveBench reasoning | 89.2% | Not measured | Only Claude Opus 4.8 measured |
| MMMU Pro (Vals AI) | 86.6% | 86.7% | GPT-5.2 |
| GPQA Diamond (Artificial Analysis run) | 92.0% | 90.3% | Claude Opus 4.8 |
| Humanity's Last Exam (Artificial Analysis run) | 48.7% | 37.7% | Claude Opus 4.8 |
| GPQA Diamond (Epoch AI run) | 91.0% | 91.4% | GPT-5.2 |
| ARC-AGI | 92.5% | 86.2% | Claude Opus 4.8 |
| ARC-AGI-2 | 72.1% | 52.9% | Claude Opus 4.8 |
| SimpleBench | 64.8% | 45.8% | Claude Opus 4.8 |
| Chess puzzles | 52.0% | 48.0% | Claude Opus 4.8 |
| Mystery game puzzles | 36.0% | 23.0% | Claude Opus 4.8 |
| EBR-bench | 28.6% | 23.0% | Claude Opus 4.8 |
| VPCT | Not measured | 84.0% | Only GPT-5.2 measured |
| LMArena Text: hard prompts | 1482 | 1428 | Claude Opus 4.8 |
| LMArena Text: expert | 1503 | 1445 | Claude Opus 4.8 |
| LMArena Text: hard prompts in English | 1487 | 1431 | Claude Opus 4.8 |
| LMArena Vision: diagrams | 1315 | 1265 | Claude Opus 4.8 |
| LMArena Vision: homework | 1337 | 1276 | Claude Opus 4.8 |
| Neither model has a result on the other 3 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| AIME | Not measured | 96.9% | Only GPT-5.2 measured |
| MGSM (Vals AI) | Not measured | 94.0% | Only GPT-5.2 measured |
| AA Math Index | Not measured | 99.0 | Only GPT-5.2 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 99.0% | Only GPT-5.2 measured |
| OTIS Mock AIME | 98.3% | 96.1% | Claude Opus 4.8 |
| FrontierMath, tiers 1 to 3 | 80.0% | 67.4% | Claude Opus 4.8 |
| FrontierMath, tier 4 | 56.1% | 31.7% | Claude Opus 4.8 |
| ProofBench | 69.0% | 15.0% | Claude Opus 4.8 |
| LMArena Text: math | 1486 | 1440 | Claude Opus 4.8 |
| LMArena Text: mathematical work | 1497 | 1442 | Claude Opus 4.8 |
| Neither model has a result on the other 3 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index101
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| MMLU Pro | 89.6% | 86.2% | Claude Opus 4.8 |
| LegalBench (Vals AI) | 83.6% | 82.8% | Claude Opus 4.8 |
| CaseLaw (Vals AI) | Not measured | 66.0% | Only GPT-5.2 measured |
| CorpFin (Vals AI) | 66.7% | 65.9% | Claude Opus 4.8 |
| TaxEval (Vals AI) | 75.6% | 75.8% | GPT-5.2 |
| MortgageTax (Vals AI) | 69.9% | 67.1% | Claude Opus 4.8 |
| MedQA (Vals AI) | Not measured | 94.1% | Only GPT-5.2 measured |
| MedCode (Vals AI) | 53.2% | 49.8% | Claude Opus 4.8 |
| MMLU Pro (Artificial Analysis run) | Not measured | 87.4% | Only GPT-5.2 measured |
| SimpleQA Verified | 53.0% | 37.1% | Claude Opus 4.8 |
| LMArena Text: science | 1477 | 1425 | Claude Opus 4.8 |
| LMArena Text: medicine and healthcare | 1464 | 1425 | Claude Opus 4.8 |
| LMArena Text: legal and government | 1477 | 1427 | Claude Opus 4.8 |
| LMArena Text: business and finance | 1458 | 1414 | Claude Opus 4.8 |
| Neither model has a result on the other 3 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index80
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| Design Arena: websites | 1258 | 1196 | Claude Opus 4.8 |
| Design Arena: UI components | 1256 | 1190 | Claude Opus 4.8 |
| Design Arena: game dev | 1264 | 1196 | Claude Opus 4.8 |
| Design Arena: data viz | 1240 | 1202 | Claude Opus 4.8 |
| Design Arena: 3D | 1223 | 1079 | Claude Opus 4.8 |
| Design Arena: all code categories | 1255 | 1171 | Claude Opus 4.8 |
| Design Arena: SVG | 1194 | 1137 | Claude Opus 4.8 |
| Design Arena: ASCII art | 1276 | 1153 | Claude Opus 4.8 |
| LiveBench design | 78.8% | Not measured | Only Claude Opus 4.8 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index292
| Benchmark | Claude Opus 4.8 | GPT-5.2 | Leader |
|---|---|---|---|
| LiveBench data analysis | 66.0% | Not measured | Only Claude Opus 4.8 measured |
| Public Benefits Bench (Vals AI) | 68.1% | Not measured | Only Claude Opus 4.8 measured |
| SAGE (Vals AI) | 54.8% | 49.3% | Claude Opus 4.8 |
| MedScribe (Vals AI) | 85.8% | 84.4% | Claude Opus 4.8 |
| Vals Multimodal Index | 70.9% | Not measured | Only Claude Opus 4.8 measured |
| IFBench | 62.2% | 75.4% | GPT-5.2 |
| Long context reasoning (AA-LCR) | 77.7% | 82.7% | GPT-5.2 |
| DTBench | 94.9% | 90.9% | Claude Opus 4.8 |
| LMCA | 57.5% | 43.9% | Claude Opus 4.8 |
| CL-bench | Not measured | 18.2% | Only GPT-5.2 measured |
| Surface Evolver Bench | 87.5% | Not measured | Only Claude Opus 4.8 measured |
| Furniture assembly | 42.5% | 38.3% | Claude Opus 4.8 |
| LMArena Text | 1461 | 1416 | Claude Opus 4.8 |
| LMArena Text: instruction following | 1476 | 1409 | Claude Opus 4.8 |
| LMArena Text: creative writing | 1454 | 1376 | Claude Opus 4.8 |
| LMArena Vision | 1294 | 1246 | Claude Opus 4.8 |
| LMArena Text: English | 1468 | 1422 | Claude Opus 4.8 |
| LMArena Text: non-English | 1450 | 1405 | Claude Opus 4.8 |
| LMArena Text: Chinese | 1506 | 1447 | Claude Opus 4.8 |
| LMArena Text: German | 1472 | 1444 | Claude Opus 4.8 |
| LMArena Text: Japanese | 1442 | 1420 | Claude Opus 4.8 |
| LMArena Text: Korean | 1431 | 1374 | Claude Opus 4.8 |
| LMArena Text: French | 1481 | 1454 | Claude Opus 4.8 |
| LMArena Text: Spanish | 1465 | 1413 | Claude Opus 4.8 |
| LMArena Text: Russian | 1474 | 1415 | Claude Opus 4.8 |
| LMArena Text: Polish | 1459 | 1426 | Claude Opus 4.8 |
| LMArena Text: multi-turn | 1476 | 1419 | Claude Opus 4.8 |
| LMArena Text: longer queries | 1483 | 1413 | Claude Opus 4.8 |
| LMArena Text: writing and language | 1461 | 1392 | Claude Opus 4.8 |
| LMArena Text: entertainment and media | 1440 | 1378 | Claude Opus 4.8 |
| LMArena Vision: English | 1293 | 1243 | Claude Opus 4.8 |
| LMArena Vision: Chinese | 1346 | 1305 | Claude Opus 4.8 |
| LMArena Vision: reading text | 1303 | 1257 | Claude Opus 4.8 |
| LMArena Vision: humor | 1282 | 1227 | Claude Opus 4.8 |
| LMArena Vision: captioning | Not measured | 1267 | Only GPT-5.2 measured |
| LMArena Vision: recognizing things | Not measured | 1232 | Only GPT-5.2 measured |
| LMArena Vision: creative writing | 1301 | 1225 | Claude Opus 4.8 |
| LMArena Document | 1475 | 1405 | Claude Opus 4.8 |
| 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 | Claude Opus 4.8 | GPT-5.2 | 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
GPT-5.2 costs less
Repository review
50K input + 3K output tokens
GPT-5.2 costs less
Agent loop
200K input + 10K output tokens
GPT-5.2 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 | Anthropic Claude Opus 4.8 | OpenAI GPT-5.2 |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | May 27, 2026 | December 10, 2025 |
| Context window | 1M | 400K |
| Input $/Mtok | $5.00 | $1.75 |
| Output $/Mtok | $25.00 | $14.00 |
| Coding index | 66.0 | 58.9 |
| Independent measurements behind it | 2 | 2 |
| Cost per solved task | $2.17 | $1.91 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Opus 4.8 better than GPT-5.2 for coding?
Pick Claude Opus 4.8 for coding: it scores 66.0 on the Vibe Coding Index against 58.9 for GPT-5.2. Pick GPT-5.2 when output tokens cost 1.8x less, or when a solved task costs $1.91 against $2.17.
Which is cheaper, Claude Opus 4.8 or GPT-5.2?
Claude Opus 4.8 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.2 costs $1.75 and $14.00. Per solved SWE-bench Verified task, Claude Opus 4.8 costs $2.17 and GPT-5.2 $1.91.
Which has the bigger context window, Claude Opus 4.8 or GPT-5.2?
Claude Opus 4.8, with 1M tokens against 400K.
How many benchmarks is the Claude Opus 4.8 vs GPT-5.2 comparison based on?
93 benchmarks have a result for both models as of October 21, 2026. Claude Opus 4.8's coding index rests on 2 independent benchmarks and GPT-5.2's on 2. Where only one model has a result, the page says so instead of naming a winner.