Claude Opus 4.6 vs gpt-oss-20b
Claude Opus 4.6 leads gpt-oss-20b on the Artificial Analysis Intelligence Index, 31.9 to 10.0, and gpt-oss-20b costs 166.7x less per output token ($0.15 vs $25.00 per million). Updated October 22, 2026; 53 benchmarks have a result for both.
Claude Opus 4.6 or gpt-oss-20b: which to pick
Only Claude Opus 4.6 has a coding score, 57.2 on the Vibe Coding Index; gpt-oss-20b has none, so the two cannot be ranked on coding.
Pick gpt-oss-20b for output tokens 166.7x cheaper ($0.15 vs $25.00 per million) and work measured by IFBench, where it leads.
- Shared results
- 53
- Claude Opus 4.6 only
- 57
- gpt-oss-20b only
- 9
- Like-for-like categories
- 8 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
Only Claude Opus 4.6 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
Claude Opus 4.6
Claude Opus 4.6 leads, 31.9 against 10.0.
Same board, both measured
Reasoning
GPQA Diamond
Claude Opus 4.6
Claude Opus 4.6 leads, 89.7% against 68.9%.
Same board, both measured
Math
AIME
Claude Opus 4.6
Claude Opus 4.6 leads, 95.6% against 86.0%.
Same board, both measured
Knowledge
MMLU Pro
Claude Opus 4.6
Claude Opus 4.6 leads, 89.1% against 71.6%.
Same board, both measured
Web design
Design Arena: websites
Claude Opus 4.6
Claude Opus 4.6 leads, 1291 against 852.
Same board, both measured
Long documents
Listed context window
Claude Opus 4.6
Claude Opus 4.6 takes 1M tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
gpt-oss-20b
Claude Opus 4.6 $0.018, gpt-oss-20b $0.0001 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Claude Opus 4.6 has the number, $1.56.
Directional only
At a glance
- Coding index
- Claude Opus 4.6 57.2 from 4 benchmarks, gpt-oss-20b not benchmarked.
- Widest category
- Web design, 49.1 points apart on average over 2 shared benchmarks.
- Widest single gap
- Terminal-Bench: Claude Opus 4.6 79.8% against gpt-oss-20b 3.4%, 76.4 points apart on a 0 to 100 scale.
- Output price
- Claude Opus 4.6 $25.00 against gpt-oss-20b $0.15 per million output tokens, 166.7x apart.
- Context window
- Claude Opus 4.6 takes 1M tokens, gpt-oss-20b 131K.
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 4.6 | Bars | gpt-oss-20b | Ties |
|---|---|---|---|---|
| Coding | 6 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 8 | 0 | ||
| Math | 4 | 0 | ||
| Knowledge | 10 | 0 | ||
| Web design | 2 | 0 | ||
| General | 19 | 1 | ||
| All shared boards | 52 | 1 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index60
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| SWE-bench Verified | 75.6% | Not measured | Only Claude Opus 4.6 measured |
| SWE-bench Multilingual | 72.0% | Not measured | Only Claude Opus 4.6 measured |
| LMArena WebDev | 1547 | Not measured | Only Claude Opus 4.6 measured |
| LiveCodeBench (Vals AI) | 84.7% | 80.4% | Claude Opus 4.6 |
| SWE-bench Verified (Vals AI) | 78.2% | Not measured | Only Claude Opus 4.6 measured |
| Vibe Code Bench (Vals AI) | 57.6% | Not measured | Only Claude Opus 4.6 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 77.7% | Only gpt-oss-20b measured |
| SciCode | Not measured | 38.9% | Only gpt-oss-20b measured |
| Terminal-Bench Hard (Artificial Analysis run) | 48.5% | 10.6% | Claude Opus 4.6 |
| Terminal-Bench | 79.8% | 3.4% | Claude Opus 4.6 |
| FrontierCode | 26.6% | Not measured | Only Claude Opus 4.6 measured |
| GSO-Bench | 41.2% | Not measured | Only Claude Opus 4.6 measured |
| WeirdML | 78.0% | 40.9% | Claude Opus 4.6 |
| Cybench | 93.0% | Not measured | Only Claude Opus 4.6 measured |
| ALE-Bench | Not measured | 566.0 | Only gpt-oss-20b measured |
| LMArena Text: coding | 1534 | 1307 | Claude Opus 4.6 |
| LMArena Text: software and IT | 1527 | 1311 | Claude Opus 4.6 |
| LMArena WebDev: React | 1539 | Not measured | Only Claude Opus 4.6 measured |
| LMArena WebDev: HTML | 1559 | Not measured | Only Claude Opus 4.6 measured |
| LMArena WebDev: from an image | 1543 | Not measured | Only Claude Opus 4.6 measured |
| Neither model has a result on the other 19 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| AA Intelligence Index | 31.9 | 10.0 | Claude Opus 4.6 |
| Epoch Capabilities Index | 155.7 | 137.7 | Claude Opus 4.6 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1206 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: agents building full-stack apps | 1220 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: agents building mobile apps | 1207 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: agents building Android apps | 1197 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: agents building games | 1212 | Not measured | Only Claude Opus 4.6 measured |
| τ²-Bench Telecom | 92.1% | 60.2% | Claude Opus 4.6 |
| APEX-Agents | 46.3% | Not measured | Only Claude Opus 4.6 measured |
| METR time horizons | 78.9% | Not measured | Only Claude Opus 4.6 measured |
| Remote Labor Index | 4.2% | Not measured | Only Claude Opus 4.6 measured |
| DeepResearch Bench | 55.3% | Not measured | Only Claude Opus 4.6 measured |
| Vending-Bench 2 | $8,018 | Not measured | Only Claude Opus 4.6 measured |
| BrowseComp-Plus | Not measured | 50.6% | Only gpt-oss-20b measured |
| Neither model has a result on the other 29 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index80
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| GPQA Diamond | 89.7% | 68.9% | Claude Opus 4.6 |
| MMMU Pro (Vals AI) | 83.9% | Not measured | Only Claude Opus 4.6 measured |
| GPQA Diamond (Artificial Analysis run) | 89.6% | 68.8% | Claude Opus 4.6 |
| Humanity's Last Exam (Artificial Analysis run) | 39.9% | 11.0% | Claude Opus 4.6 |
| GPQA Diamond (Epoch AI run) | 90.5% | 60.8% | Claude Opus 4.6 |
| ARC-AGI | 94.0% | Not measured | Only Claude Opus 4.6 measured |
| ARC-AGI-2 | 69.2% | Not measured | Only Claude Opus 4.6 measured |
| SimpleBench | 67.6% | Not measured | Only Claude Opus 4.6 measured |
| Chess puzzles | 21.0% | Not measured | Only Claude Opus 4.6 measured |
| Mystery game puzzles | 25.0% | Not measured | Only Claude Opus 4.6 measured |
| EBR-bench | 12.7% | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text: hard prompts | 1525 | 1274 | Claude Opus 4.6 |
| LMArena Text: expert | 1544 | 1258 | Claude Opus 4.6 |
| LMArena Text: hard prompts in English | 1530 | 1292 | Claude Opus 4.6 |
| LMArena Vision: diagrams | 1340 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: homework | 1337 | Not measured | Only Claude Opus 4.6 measured |
| Kagi LLM Benchmark | 83.6% | 53.2% | Claude Opus 4.6 |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| AIME | 95.6% | 86.0% | Claude Opus 4.6 |
| MATH 500 (Vals AI) | Not measured | 94.2% | Only gpt-oss-20b measured |
| MGSM (Vals AI) | Not measured | 89.0% | Only gpt-oss-20b measured |
| AA Math Index | Not measured | 89.3 | Only gpt-oss-20b measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 89.3% | Only gpt-oss-20b measured |
| OTIS Mock AIME | 94.4% | 65.3% | Claude Opus 4.6 |
| FrontierMath, tiers 1 to 3 | 66.0% | Not measured | Only Claude Opus 4.6 measured |
| FrontierMath, tier 4 | 26.8% | Not measured | Only Claude Opus 4.6 measured |
| ProofBench | 50.0% | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text: math | 1518 | 1317 | Claude Opus 4.6 |
| LMArena Text: mathematical work | 1525 | 1305 | Claude Opus 4.6 |
| Neither model has a result on the other 2 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index100
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| MMLU Pro | 89.1% | 71.6% | Claude Opus 4.6 |
| LegalBench (Vals AI) | 85.3% | 70.8% | Claude Opus 4.6 |
| CaseLaw (Vals AI) | 62.1% | 43.8% | Claude Opus 4.6 |
| CorpFin (Vals AI) | 67.0% | 53.1% | Claude Opus 4.6 |
| TaxEval (Vals AI) | 76.0% | 63.7% | Claude Opus 4.6 |
| MortgageTax (Vals AI) | 68.5% | Not measured | Only Claude Opus 4.6 measured |
| MedQA (Vals AI) | 95.4% | 82.9% | Claude Opus 4.6 |
| MedCode (Vals AI) | 49.1% | Not measured | Only Claude Opus 4.6 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 74.8% | Only gpt-oss-20b measured |
| SimpleQA Verified | 47.0% | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text: science | 1519 | 1296 | Claude Opus 4.6 |
| LMArena Text: medicine and healthcare | 1504 | 1310 | Claude Opus 4.6 |
| LMArena Text: legal and government | 1512 | 1280 | Claude Opus 4.6 |
| LMArena Text: business and finance | 1497 | 1284 | Claude Opus 4.6 |
| FACTS Benchmark Suite | 41.5% | Not measured | Only Claude Opus 4.6 measured |
| Neither model has a result on the other 3 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| Design Arena: websites | 1291 | 852 | Claude Opus 4.6 |
| Design Arena: UI components | 1281 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: game dev | 1277 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: data viz | 1279 | 933 | Claude Opus 4.6 |
| Design Arena: 3D | 1273 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: all code categories | 1288 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: SVG | 1227 | Not measured | Only Claude Opus 4.6 measured |
| Design Arena: ASCII art | 1255 | Not measured | Only Claude Opus 4.6 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index191
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 51.6% | Not measured | Only Claude Opus 4.6 measured |
| MedScribe (Vals AI) | 86.7% | Not measured | Only Claude Opus 4.6 measured |
| IFBench | 53.1% | 65.1% | gpt-oss-20b |
| Long context reasoning (AA-LCR) | 78.0% | 34.7% | Claude Opus 4.6 |
| DTBench | 91.2% | 68.0% | Claude Opus 4.6 |
| LMCA | 55.8% | 14.5% | Claude Opus 4.6 |
| CL-bench | 20.7% | Not measured | Only Claude Opus 4.6 measured |
| Furniture assembly | 28.3% | Not measured | Only Claude Opus 4.6 measured |
| CL-bench Life | 17.0% | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text | 1501 | 1288 | Claude Opus 4.6 |
| LMArena Text: instruction following | 1521 | 1236 | Claude Opus 4.6 |
| LMArena Text: creative writing | 1505 | 1201 | Claude Opus 4.6 |
| LMArena Vision | 1316 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text: English | 1511 | 1303 | Claude Opus 4.6 |
| LMArena Text: non-English | 1488 | 1269 | Claude Opus 4.6 |
| LMArena Text: Chinese | 1551 | 1315 | Claude Opus 4.6 |
| LMArena Text: German | 1500 | 1255 | Claude Opus 4.6 |
| LMArena Text: Japanese | 1488 | 1244 | Claude Opus 4.6 |
| LMArena Text: Korean | 1464 | 1236 | Claude Opus 4.6 |
| LMArena Text: French | 1512 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Text: Spanish | 1510 | 1267 | Claude Opus 4.6 |
| LMArena Text: Russian | 1497 | 1278 | Claude Opus 4.6 |
| LMArena Text: Polish | 1499 | 1252 | Claude Opus 4.6 |
| LMArena Text: multi-turn | 1511 | 1268 | Claude Opus 4.6 |
| LMArena Text: longer queries | 1518 | 1250 | Claude Opus 4.6 |
| LMArena Text: writing and language | 1502 | 1238 | Claude Opus 4.6 |
| LMArena Text: entertainment and media | 1492 | 1216 | Claude Opus 4.6 |
| LMArena Vision: English | 1324 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: Chinese | 1370 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: reading text | 1326 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: humor | 1314 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: recognizing things | 1274 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Vision: creative writing | 1331 | Not measured | Only Claude Opus 4.6 measured |
| LMArena Document | 1507 | Not measured | Only Claude Opus 4.6 measured |
| Neither model has a result on the other 12 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Claude Opus 4.6 | gpt-oss-20b | 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-oss-20b costs less
Repository review
50K input + 3K output tokens
gpt-oss-20b costs less
Agent loop
200K input + 10K output tokens
gpt-oss-20b 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 4.6 | OpenAI gpt-oss-20b |
|---|---|---|
| Maker | Anthropic | OpenAI |
| Weights | No public weights listed | Open weights |
| Released | February 4, 2026 | August 5, 2025 |
| Context window | 1M | 131K |
| Input $/Mtok | $5.00 | $0.03 |
| Output $/Mtok | $25.00 | $0.15 |
| Coding index | 57.2 | Not benchmarked |
| Independent measurements behind it | 4 | 0 |
| Cost per solved task | $1.56 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Opus 4.6 better than gpt-oss-20b for coding?
Only Claude Opus 4.6 has a coding score, 57.2 on the Vibe Coding Index; gpt-oss-20b has none, so the two cannot be ranked on coding. Pick gpt-oss-20b for output tokens 166.7x cheaper ($0.15 vs $25.00 per million) and work measured by IFBench, where it leads.
Which is cheaper, Claude Opus 4.6 or gpt-oss-20b?
Claude Opus 4.6 costs $5.00 per million input tokens and $25.00 per million output tokens; gpt-oss-20b costs $0.03 and $0.15. Cost per solved task is known for Claude Opus 4.6 only.
Which has the bigger context window, Claude Opus 4.6 or gpt-oss-20b?
Claude Opus 4.6, with 1M tokens against 131K.
How many benchmarks is the Claude Opus 4.6 vs gpt-oss-20b comparison based on?
53 benchmarks have a result for both models as of October 22, 2026. Claude Opus 4.6's coding index rests on 4 independent benchmarks and gpt-oss-20b's on 0. Where only one model has a result, the page says so instead of naming a winner.