Claude Opus 4.5 vs GPT-5.2-Codex
GPT-5.2-Codex leads Claude Opus 4.5 on coding, 45.4 to 44.0 on the Vibe Coding Index, and GPT-5.2-Codex costs 1.8x less per output token ($14.00 vs $25.00 per million). Updated October 21, 2026; 20 benchmarks have a result for both.
Claude Opus 4.5 or GPT-5.2-Codex: which to pick
Claude Opus 4.5 and GPT-5.2-Codex are level for coding, 44.0 against 45.4 on the Vibe Coding Index, so price and context decide.
Pick Claude Opus 4.5 for work measured by SWE-bench Verified or SWE-bench Multilingual, where it leads. Pick GPT-5.2-Codex for output tokens 1.8x cheaper ($14.00 vs $25.00 per million) and a lower cost per solved task ($0.43 against $1.37).
- Shared results
- 20
- Claude Opus 4.5 only
- 88
- GPT-5.2-Codex only
- 2
- Like-for-like categories
- 5 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
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
Level: 44.0 against 45.4, inside 2 points.
3 benchmarks behind each score
Intelligence
AA Intelligence Index
Claude Opus 4.5
Claude Opus 4.5 leads, 29.1 against 28.5.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only Claude Opus 4.5 has a GPQA Diamond result, 85.9%.
Directional only
Math
AIME
No call
Only Claude Opus 4.5 has a AIME result, 95.4%.
Directional only
Knowledge
MMLU Pro
No call
Only Claude Opus 4.5 has a MMLU Pro result, 87.3%.
Directional only
Web design
Design Arena: websites
No call
Only Claude Opus 4.5 has a Design Arena: websites result, 1247.
Directional only
Long documents
Listed context window
GPT-5.2-Codex
GPT-5.2-Codex takes 400K tokens against 200K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.2-Codex
Claude Opus 4.5 $0.018, GPT-5.2-Codex $0.0088 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.2-Codex
Claude Opus 4.5 $1.37, GPT-5.2-Codex $0.43.
Published attempt cost and pass rate
At a glance
- Coding index
- Claude Opus 4.5 44.0 from 3 benchmarks, GPT-5.2-Codex 45.4 from 3 benchmarks.
- Widest category
- Coding, 12.4 points apart on average over 10 shared benchmarks.
- Widest single gap
- LMArena WebDev: HTML: Claude Opus 4.5 1510 against GPT-5.2-Codex 1348, 27.0 points apart on a 0 to 100 scale.
- Output price
- Claude Opus 4.5 $25.00 against GPT-5.2-Codex $14.00 per million output tokens, 1.8x apart.
- Context window
- Claude Opus 4.5 takes 200K tokens, GPT-5.2-Codex 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 4.5 | Bars | GPT-5.2-Codex | Ties |
|---|---|---|---|---|
| Coding | 7 | 3 | ||
| Intelligence | 1 | 0 | ||
| Agentic work | 3 | 2 | ||
| Reasoning | 0 | 2 | ||
| General | 0 | 2 | ||
| All shared boards | 11 | 9 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index73
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| SWE-bench Verified | 79.2% | 72.8% | Claude Opus 4.5 |
| SWE-bench (full test set) | 52.6% | Not measured | Only Claude Opus 4.5 measured |
| SWE-bench Multilingual | 70.7% | 66.3% | Claude Opus 4.5 |
| LMArena WebDev | 1494 | 1339 | Claude Opus 4.5 |
| LiveCodeBench (Vals AI) | 83.7% | 88.0% | GPT-5.2-Codex |
| SWE-bench Verified (Vals AI) | 76.4% | 72.4% | Claude Opus 4.5 |
| Vibe Code Bench (Vals AI) | 20.6% | 37.9% | GPT-5.2-Codex |
| LiveCodeBench (Artificial Analysis run) | 87.1% | Not measured | Only Claude Opus 4.5 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 47.0% | 37.1% | Claude Opus 4.5 |
| Terminal-Bench | 63.1% | 66.5% | GPT-5.2-Codex |
| GSO-Bench | 26.5% | Not measured | Only Claude Opus 4.5 measured |
| WeirdML | 63.7% | Not measured | Only Claude Opus 4.5 measured |
| Cybench | 82.0% | Not measured | Only Claude Opus 4.5 measured |
| ALE-Bench | Not measured | 1299.9 | Only GPT-5.2-Codex measured |
| LMArena Text: coding | 1503 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: software and IT | 1483 | Not measured | Only Claude Opus 4.5 measured |
| LMArena WebDev: React | 1478 | 1329 | Claude Opus 4.5 |
| LMArena WebDev: HTML | 1510 | 1348 | Claude Opus 4.5 |
| Neither model has a result on the other 21 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| AA Intelligence Index | 29.1 | 28.5 | Claude Opus 4.5 |
| Epoch Capabilities Index | 150.1 | Not measured | Only Claude Opus 4.5 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index32
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1159 | 1051 | Claude Opus 4.5 |
| Design Arena: agents building full-stack apps | 1156 | 993 | Claude Opus 4.5 |
| Design Arena: agents building mobile apps | 1189 | 1090 | Claude Opus 4.5 |
| Design Arena: agents building Android apps | 1175 | 1176 | GPT-5.2-Codex |
| Design Arena: agents building Godot games | Not measured | 1142 | Only GPT-5.2-Codex measured |
| τ²-Bench Telecom | 89.5% | 92.1% | GPT-5.2-Codex |
| METR time horizons | 75.0% | Not measured | Only Claude Opus 4.5 measured |
| Remote Labor Index | 3.8% | Not measured | Only Claude Opus 4.5 measured |
| GDPval | 45.5% | Not measured | Only Claude Opus 4.5 measured |
| BALROG | 43.5% | Not measured | Only Claude Opus 4.5 measured |
| DeepResearch Bench | 54.8% | Not measured | Only Claude Opus 4.5 measured |
| OSWorld | 66.3% | Not measured | Only Claude Opus 4.5 measured |
| Vending-Bench 2 | $4,967 | Not measured | Only Claude Opus 4.5 measured |
| Berkeley Function Calling Leaderboard (BFCL) | 77.5% | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 27 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| GPQA Diamond | 85.9% | Not measured | Only Claude Opus 4.5 measured |
| MMMU Pro (Vals AI) | 83.0% | Not measured | Only Claude Opus 4.5 measured |
| GPQA Diamond (Artificial Analysis run) | 86.6% | 89.9% | GPT-5.2-Codex |
| Humanity's Last Exam (Artificial Analysis run) | 30.1% | 35.7% | GPT-5.2-Codex |
| GPQA Diamond (Epoch AI run) | 86.0% | Not measured | Only Claude Opus 4.5 measured |
| ARC-AGI | 80.0% | Not measured | Only Claude Opus 4.5 measured |
| ARC-AGI-2 | 37.6% | Not measured | Only Claude Opus 4.5 measured |
| SimpleBench | 62.0% | Not measured | Only Claude Opus 4.5 measured |
| Chess puzzles | 14.0% | Not measured | Only Claude Opus 4.5 measured |
| Mystery game puzzles | 22.0% | Not measured | Only Claude Opus 4.5 measured |
| EBR-bench | 14.3% | Not measured | Only Claude Opus 4.5 measured |
| VPCT | 40.0% | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: hard prompts | 1476 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: expert | 1487 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: hard prompts in English | 1482 | Not measured | Only Claude Opus 4.5 measured |
| Kagi LLM Benchmark | 80.2% | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 6 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index10 boards, none shared
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| AIME | 95.4% | Not measured | Only Claude Opus 4.5 measured |
| MGSM (Vals AI) | 95.2% | Not measured | Only Claude Opus 4.5 measured |
| AA Math Index | 91.3 | Not measured | Only Claude Opus 4.5 measured |
| AIME 2025 (Artificial Analysis run) | 91.3% | Not measured | Only Claude Opus 4.5 measured |
| OTIS Mock AIME | 86.1% | Not measured | Only Claude Opus 4.5 measured |
| FrontierMath, tiers 1 to 3 | 34.4% | Not measured | Only Claude Opus 4.5 measured |
| FrontierMath, tier 4 | 4.9% | Not measured | Only Claude Opus 4.5 measured |
| ProofBench | 36.0% | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: math | 1463 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: mathematical work | 1468 | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 3 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index16 boards, none shared
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| MMLU Pro | 87.3% | Not measured | Only Claude Opus 4.5 measured |
| LegalBench (Vals AI) | 84.6% | Not measured | Only Claude Opus 4.5 measured |
| CaseLaw (Vals AI) | 62.6% | Not measured | Only Claude Opus 4.5 measured |
| CorpFin (Vals AI) | 65.1% | Not measured | Only Claude Opus 4.5 measured |
| TaxEval (Vals AI) | 74.9% | Not measured | Only Claude Opus 4.5 measured |
| MortgageTax (Vals AI) | 68.7% | Not measured | Only Claude Opus 4.5 measured |
| MedQA (Vals AI) | 95.9% | Not measured | Only Claude Opus 4.5 measured |
| MedCode (Vals AI) | 49.2% | Not measured | Only Claude Opus 4.5 measured |
| MMLU Pro (Artificial Analysis run) | 89.5% | Not measured | Only Claude Opus 4.5 measured |
| SimpleQA Verified | 45.7% | Not measured | Only Claude Opus 4.5 measured |
| GeoBench | 75.0% | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: science | 1454 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: medicine and healthcare | 1444 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: legal and government | 1459 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: business and finance | 1448 | Not measured | Only Claude Opus 4.5 measured |
| FACTS Benchmark Suite | 43.9% | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 2 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| Design Arena: websites | 1247 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: UI components | 1238 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: game dev | 1229 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: data viz | 1243 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: 3D | 1214 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: all code categories | 1242 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: SVG | 1181 | Not measured | Only Claude Opus 4.5 measured |
| Design Arena: ASCII art | 1190 | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 52.1% | Not measured | Only Claude Opus 4.5 measured |
| MedScribe (Vals AI) | 85.3% | Not measured | Only Claude Opus 4.5 measured |
| IFBench | 58.0% | 77.6% | GPT-5.2-Codex |
| Long context reasoning (AA-LCR) | 77.3% | 82.3% | GPT-5.2-Codex |
| DTBench | 89.9% | Not measured | Only Claude Opus 4.5 measured |
| LMCA | 44.5% | Not measured | Only Claude Opus 4.5 measured |
| CL-bench | 21.1% | Not measured | Only Claude Opus 4.5 measured |
| Furniture assembly | 28.3% | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text | 1451 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: instruction following | 1478 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: creative writing | 1445 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: English | 1461 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: non-English | 1438 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Chinese | 1470 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: German | 1449 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Japanese | 1417 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Korean | 1424 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: French | 1471 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Spanish | 1457 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Russian | 1447 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: Polish | 1439 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: multi-turn | 1466 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: longer queries | 1479 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: writing and language | 1452 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Text: entertainment and media | 1434 | Not measured | Only Claude Opus 4.5 measured |
| LMArena Document | 1462 | Not measured | Only Claude Opus 4.5 measured |
| Neither model has a result on the other 20 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Claude Opus 4.5 | GPT-5.2-Codex | 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-Codex costs less
Repository review
50K input + 3K output tokens
GPT-5.2-Codex costs less
Agent loop
200K input + 10K output tokens
GPT-5.2-Codex 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 4.5 | OpenAI GPT-5.2-Codex |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Weights | No public weights listed | No public weights listed |
| Released | November 24, 2025 | January 14, 2026 |
| Context window | 200K | 400K |
| Input $/Mtok | $5.00 | $1.75 |
| Output $/Mtok | $25.00 | $14.00 |
| Coding index | 44.0 | 45.4 |
| Independent measurements behind it | 3 | 3 |
| Cost per solved task | $1.37 | $0.43 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Opus 4.5 better than GPT-5.2-Codex for coding?
Claude Opus 4.5 and GPT-5.2-Codex are level for coding, 44.0 against 45.4 on the Vibe Coding Index, so price and context decide. Pick Claude Opus 4.5 for work measured by SWE-bench Verified or SWE-bench Multilingual, where it leads. Pick GPT-5.2-Codex for output tokens 1.8x cheaper ($14.00 vs $25.00 per million) and a lower cost per solved task ($0.43 against $1.37).
Which is cheaper, Claude Opus 4.5 or GPT-5.2-Codex?
Claude Opus 4.5 costs $5.00 per million input tokens and $25.00 per million output tokens; GPT-5.2-Codex costs $1.75 and $14.00. Per solved SWE-bench Verified task, Claude Opus 4.5 costs $1.37 and GPT-5.2-Codex $0.43.
Which has the bigger context window, Claude Opus 4.5 or GPT-5.2-Codex?
GPT-5.2-Codex, with 400K tokens against 200K.
How many benchmarks is the Claude Opus 4.5 vs GPT-5.2-Codex comparison based on?
20 benchmarks have a result for both models as of October 21, 2026. Claude Opus 4.5's coding index rests on 3 independent benchmarks and GPT-5.2-Codex's on 3. Where only one model has a result, the page says so instead of naming a winner.