GPT-5.6 Terra vs Qwen3.8 Max (0902)
GPT-5.6 Terra leads Qwen3.8 Max (0902) on coding, 63.9 to 62.4 on the Vibe Coding Index, and Qwen3.8 Max (0902) costs 2.0x less per output token ($6.00 vs $12.00 per million). Updated October 22, 2026; 107 benchmarks have a result for both.
GPT-5.6 Terra or Qwen3.8 Max (0902): which to pick
GPT-5.6 Terra and Qwen3.8 Max (0902) are level for coding, 63.9 against 62.4 on the Vibe Coding Index, so price and context decide.
Pick Qwen3.8 Max (0902) for output tokens 2.0x cheaper ($6.00 vs $12.00 per million) and work measured by LMArena WebDev or LiveCodeBench (Vals AI), where it leads. Pick GPT-5.6 Terra for a lower cost per solved task ($0.42 against $1.32) and a larger context window (1.05M against 1M tokens).
- Shared results
- 107
- GPT-5.6 Terra only
- 22
- Qwen3.8 Max (0902) only
- 7
- 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
Level: 63.9 against 62.4, inside 2 points.
6 and 5 benchmarks behind the scores
Intelligence
AA Intelligence Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 45.4 against 42.1.
Same board, both measured
Agentic work
AA Agentic Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 56.0 against 43.2.
Same board, both measured
Reasoning
GPQA Diamond
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 93.7% against 90.9%.
Same board, both measured
Knowledge
MMLU Pro
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 88.6% against 86.7%.
Same board, both measured
Web design
Design Arena: websites
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 1275 against 1268.
Same board, both measured
Long documents
Listed context window
GPT-5.6 Terra
GPT-5.6 Terra takes 1.05M tokens against 1M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3.8 Max (0902)
GPT-5.6 Terra $0.0080, Qwen3.8 Max (0902) $0.0050 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-5.6 Terra
GPT-5.6 Terra $0.42, Qwen3.8 Max (0902) $1.32.
Published attempt cost and pass rate
At a glance
- Coding index
- GPT-5.6 Terra 63.9 from 6 benchmarks, Qwen3.8 Max (0902) 62.4 from 5 benchmarks.
- Widest category
- Agentic work, 13.7 points apart on average over 15 shared benchmarks.
- Widest single gap
- LiveBench browser: GPT-5.6 Terra 84.0% against Qwen3.8 Max (0902) 32.0%, 52.0 points apart on a 0 to 100 scale.
- Output price
- GPT-5.6 Terra $12.00 against Qwen3.8 Max (0902) $6.00 per million output tokens, 2.0x apart.
- Context window
- GPT-5.6 Terra takes 1.05M tokens, Qwen3.8 Max (0902) 1M.
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 | GPT-5.6 Terra | Bars | Qwen3.8 Max (0902) | Ties |
|---|---|---|---|---|
| Coding | 12 | 8 | ||
| Intelligence | 1 | 1 | ||
| Agentic work | 4 | 11 | ||
| Reasoning | 3 | 10 | ||
| Math | 3 | 3 | ||
| Knowledge | 4 | 7 | ||
| Web design | 0 | 8 | ||
| General | 5 | 27 | ||
| All shared boards | 32 | 75 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index128
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| LiveBench coding | 86.2% | 60.3% | GPT-5.6 Terra |
| LMArena WebDev | 1522 | 1674 | Qwen3.8 Max (0902) |
| AA Coding Index | 76.7 | 76.2 | GPT-5.6 Terra |
| LiveCodeBench (Vals AI) | 85.9% | 87.8% | Qwen3.8 Max (0902) |
| SWE-bench Verified (Vals AI) | 95.4% | 85.6% | GPT-5.6 Terra |
| Terminal-Bench 4 (Vals AI) | 22.7% | 34.3% | Qwen3.8 Max (0902) |
| Vibe Code Bench (Vals AI) | 74.6% | 64.7% | GPT-5.6 Terra |
| Code migration (Vals AI) | 47.8% | 24.0% | GPT-5.6 Terra |
| ProgramBench (Vals AI) | 0.5% | 0.0% | GPT-5.6 Terra |
| IOI (Vals AI) | 87.6% | 68.9% | GPT-5.6 Terra |
| Terminal-Bench 2.1 (Vals AI) | 77.5% | 67.4% | GPT-5.6 Terra |
| SWE-bench Multimodal (Vals AI) | Not measured | 42.5% | Only Qwen3.8 Max (0902) measured |
| Vibe Code Bench 1-100 (Vals AI) | 14.8% | 12.8% | GPT-5.6 Terra |
| SkillsBench (Vals AI) | 58.9% | 42.0% | GPT-5.6 Terra |
| CyberBench (Vals AI) | 72.1% | 28.6% | GPT-5.6 Terra |
| SciCode | 55.0% | 53.2% | GPT-5.6 Terra |
| Terminal-Bench Hard (Artificial Analysis run) | 62.9% | Not measured | Only GPT-5.6 Terra measured |
| FrontierCode | 41.3% | Not measured | Only GPT-5.6 Terra measured |
| FrontierSWE | Not measured | 17.8% | Only Qwen3.8 Max (0902) measured |
| WeirdML | 78.3% | Not measured | Only GPT-5.6 Terra measured |
| ALE-Bench | 1951.4 | Not measured | Only GPT-5.6 Terra measured |
| LMArena Text: coding | 1485 | 1502 | Qwen3.8 Max (0902) |
| LMArena Text: software and IT | 1482 | 1498 | Qwen3.8 Max (0902) |
| LMArena WebDev: React | 1513 | 1692 | Qwen3.8 Max (0902) |
| LMArena WebDev: HTML | 1526 | 1649 | Qwen3.8 Max (0902) |
| LMArena WebDev: from an image | 1541 | 1637 | Qwen3.8 Max (0902) |
| SciCode | 10.8% | Not measured | Only GPT-5.6 Terra measured |
| SWE-Bench Pro V2 | 92.4% | Not measured | Only GPT-5.6 Terra measured |
| Neither model has a result on the other 15 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index11
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Intelligence Index | 42.1 | 45.4 | Qwen3.8 Max (0902) |
| Epoch Capabilities Index | 160.1 | 157.4 | GPT-5.6 Terra |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index411
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Agentic Index | 43.2 | 56.0 | Qwen3.8 Max (0902) |
| Design Arena: agents building web apps | 1120 | 1300 | Qwen3.8 Max (0902) |
| Design Arena: agents building full-stack apps | 1061 | 1305 | Qwen3.8 Max (0902) |
| Design Arena: agents building mobile apps | 1110 | 1224 | Qwen3.8 Max (0902) |
| Design Arena: agents building Android apps | 1060 | 1190 | Qwen3.8 Max (0902) |
| Design Arena: agents building Godot games | Not measured | 1234 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building games | 1083 | Not measured | Only GPT-5.6 Terra measured |
| Design Arena: HTML slides | 1047 | Not measured | Only GPT-5.6 Terra measured |
| Design Arena: PowerPoint slides | 1104 | 1232 | Qwen3.8 Max (0902) |
| Design Arena: spreadsheets | Not measured | 993 | Only Qwen3.8 Max (0902) measured |
| LiveBench research | 60.4% | 69.9% | Qwen3.8 Max (0902) |
| LiveBench browser | 84.0% | 32.0% | GPT-5.6 Terra |
| LiveBench long agentic | 48.4% | 42.3% | GPT-5.6 Terra |
| SREBench (Vals AI) | 9.5% | Not measured | Only GPT-5.6 Terra measured |
| Finance Agent (Vals AI) | 54.4% | 50.6% | GPT-5.6 Terra |
| Tax Agent Bench (Vals AI) | 30.0% | 32.1% | Qwen3.8 Max (0902) |
| Legal Research Bench (Vals AI) | 41.4% | 47.6% | Qwen3.8 Max (0902) |
| Excel Modeling Benchmark (Vals AI) | 66.2% | 60.1% | GPT-5.6 Terra |
| Terminal-Bench-Science (Vals AI) | 10.0% | 12.9% | Qwen3.8 Max (0902) |
| τ²-Bench Telecom | 86.3% | Not measured | Only GPT-5.6 Terra measured |
| APEX-Agents | 58.2% | 63.3% | Qwen3.8 Max (0902) |
| BALROG | 53.2% | Not measured | Only GPT-5.6 Terra measured |
| Vending-Bench 2 | $7,343 | Not measured | Only GPT-5.6 Terra measured |
| LMArena Search | Not measured | 1261 | Only Qwen3.8 Max (0902) measured |
| Neither model has a result on the other 18 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index310
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| GPQA Diamond | 90.9% | 93.7% | Qwen3.8 Max (0902) |
| LiveBench reasoning | 90.6% | 88.2% | GPT-5.6 Terra |
| MysteryMechanism (Vals AI) | Not measured | 23.9% | Only Qwen3.8 Max (0902) measured |
| MMMU Pro (Vals AI) | 86.5% | 88.0% | Qwen3.8 Max (0902) |
| GPQA Diamond (Artificial Analysis run) | 92.5% | 92.8% | Qwen3.8 Max (0902) |
| Humanity's Last Exam (Artificial Analysis run) | 42.9% | 43.1% | Qwen3.8 Max (0902) |
| GPQA Diamond (Epoch AI run) | 93.3% | 92.7% | GPT-5.6 Terra |
| ARC-AGI | 96.5% | Not measured | Only GPT-5.6 Terra measured |
| ARC-AGI-2 | 83.9% | Not measured | Only GPT-5.6 Terra measured |
| SimpleBench | 48.9% | Not measured | Only GPT-5.6 Terra measured |
| Chess puzzles | 47.0% | 46.0% | GPT-5.6 Terra |
| Mystery game puzzles | 35.0% | 38.0% | Qwen3.8 Max (0902) |
| LMArena Text: hard prompts | 1469 | 1496 | Qwen3.8 Max (0902) |
| LMArena Text: expert | 1496 | 1507 | Qwen3.8 Max (0902) |
| LMArena Text: hard prompts in English | 1465 | 1503 | Qwen3.8 Max (0902) |
| LMArena Vision: diagrams | 1295 | 1323 | Qwen3.8 Max (0902) |
| LMArena Vision: homework | 1312 | 1336 | Qwen3.8 Max (0902) |
| Kagi LLM Benchmark | 51.3% | Not measured | Only GPT-5.6 Terra measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index33
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| OTIS Mock AIME | 99.7% | 100.0% | Qwen3.8 Max (0902) |
| FrontierMath, tiers 1 to 3 | 86.0% | 74.7% | GPT-5.6 Terra |
| FrontierMath, tier 4 | 70.7% | 46.3% | GPT-5.6 Terra |
| ProofBench | 74.0% | 58.0% | GPT-5.6 Terra |
| LMArena Text: math | 1465 | 1501 | Qwen3.8 Max (0902) |
| LMArena Text: mathematical work | 1478 | 1491 | Qwen3.8 Max (0902) |
| MathArena | Not measured | 51.6% | Only Qwen3.8 Max (0902) measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index47
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| MMLU Pro | 86.7% | 88.6% | Qwen3.8 Max (0902) |
| LegalBench (Vals AI) | 85.1% | 83.6% | GPT-5.6 Terra |
| CorpFin (Vals AI) | 65.3% | 65.8% | Qwen3.8 Max (0902) |
| TaxEval (Vals AI) | 76.2% | 75.5% | GPT-5.6 Terra |
| MortgageTax (Vals AI) | 67.3% | 64.0% | GPT-5.6 Terra |
| MedCode (Vals AI) | 43.4% | 40.7% | GPT-5.6 Terra |
| SimpleQA Verified | 43.2% | 47.3% | Qwen3.8 Max (0902) |
| LMArena Text: science | 1452 | 1512 | Qwen3.8 Max (0902) |
| LMArena Text: medicine and healthcare | 1430 | 1503 | Qwen3.8 Max (0902) |
| LMArena Text: legal and government | 1461 | 1492 | Qwen3.8 Max (0902) |
| LMArena Text: business and finance | 1448 | 1480 | Qwen3.8 Max (0902) |
| FACTS Benchmark Suite | 52.0% | Not measured | Only GPT-5.6 Terra measured |
| Neither model has a result on the other 8 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| Design Arena: websites | 1268 | 1275 | Qwen3.8 Max (0902) |
| Design Arena: UI components | 1256 | 1336 | Qwen3.8 Max (0902) |
| Design Arena: game dev | 1229 | 1312 | Qwen3.8 Max (0902) |
| Design Arena: data viz | 1240 | 1282 | Qwen3.8 Max (0902) |
| Design Arena: 3D | 1134 | 1336 | Qwen3.8 Max (0902) |
| Design Arena: all code categories | 1240 | 1292 | Qwen3.8 Max (0902) |
| Design Arena: SVG | 1219 | Not measured | Only GPT-5.6 Terra measured |
| Design Arena: ASCII art | 1178 | 1200 | Qwen3.8 Max (0902) |
| LiveBench design | 76.0% | 78.6% | Qwen3.8 Max (0902) |
GeneralLike-for-likePublic boards, shown as published, kept out of the index527
| Benchmark | GPT-5.6 Terra | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| LiveBench data analysis | 79.3% | 78.4% | GPT-5.6 Terra |
| Public Benefits Bench (Vals AI) | 62.4% | 67.1% | Qwen3.8 Max (0902) |
| SAGE (Vals AI) | 47.0% | 51.2% | Qwen3.8 Max (0902) |
| MedScribe (Vals AI) | 82.9% | 85.0% | Qwen3.8 Max (0902) |
| Vals Multimodal Index | 65.1% | 65.4% | Qwen3.8 Max (0902) |
| IFBench | 71.2% | Not measured | Only GPT-5.6 Terra measured |
| Long context reasoning (AA-LCR) | 83.0% | 80.3% | GPT-5.6 Terra |
| DTBench | 93.3% | 92.0% | GPT-5.6 Terra |
| LMCA | 54.9% | 46.2% | GPT-5.6 Terra |
| Surface Evolver Bench | 83.8% | Not measured | Only GPT-5.6 Terra measured |
| Furniture assembly | 54.2% | 20.0% | GPT-5.6 Terra |
| LMArena Text | 1448 | 1483 | Qwen3.8 Max (0902) |
| LMArena Text: instruction following | 1456 | 1481 | Qwen3.8 Max (0902) |
| LMArena Text: creative writing | 1413 | 1477 | Qwen3.8 Max (0902) |
| LMArena Vision | 1271 | 1313 | Qwen3.8 Max (0902) |
| LMArena Text: English | 1449 | 1491 | Qwen3.8 Max (0902) |
| LMArena Text: non-English | 1441 | 1473 | Qwen3.8 Max (0902) |
| LMArena Text: Chinese | 1514 | 1539 | Qwen3.8 Max (0902) |
| LMArena Text: German | 1460 | 1478 | Qwen3.8 Max (0902) |
| LMArena Text: Japanese | 1454 | 1473 | Qwen3.8 Max (0902) |
| LMArena Text: Korean | 1427 | 1460 | Qwen3.8 Max (0902) |
| LMArena Text: French | 1466 | 1498 | Qwen3.8 Max (0902) |
| LMArena Text: Spanish | 1451 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: Russian | 1452 | 1482 | Qwen3.8 Max (0902) |
| LMArena Text: Polish | 1446 | 1471 | Qwen3.8 Max (0902) |
| LMArena Text: multi-turn | 1453 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: longer queries | 1454 | 1488 | Qwen3.8 Max (0902) |
| LMArena Text: writing and language | 1431 | 1477 | Qwen3.8 Max (0902) |
| LMArena Text: entertainment and media | 1418 | 1464 | Qwen3.8 Max (0902) |
| LMArena Vision: English | 1272 | 1311 | Qwen3.8 Max (0902) |
| LMArena Vision: Chinese | 1337 | 1358 | Qwen3.8 Max (0902) |
| LMArena Vision: reading text | 1278 | 1322 | Qwen3.8 Max (0902) |
| LMArena Vision: humor | 1236 | 1304 | Qwen3.8 Max (0902) |
| LMArena Vision: creative writing | 1248 | 1312 | Qwen3.8 Max (0902) |
| LMArena Document | 1472 | Not measured | Only GPT-5.6 Terra measured |
| EQ-Bench 4 | 1234 | Not measured | Only GPT-5.6 Terra 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 | GPT-5.6 Terra | Qwen3.8 Max (0902) | 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
Qwen3.8 Max (0902) costs less
Repository review
50K input + 3K output tokens
Qwen3.8 Max (0902) costs less
Agent loop
200K input + 10K output tokens
Qwen3.8 Max (0902) 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 | OpenAI GPT-5.6 Terra | Qwen3.8 Max (0902) |
|---|---|---|
| Maker | OpenAI | Qwen |
| Weights | No public weights listed | No public weights listed |
| Released | July 9, 2026 | September 3, 2026 |
| Context window | 1.05M | 1M |
| Input $/Mtok | $2.00 | $2.00 |
| Output $/Mtok | $12.00 | $6.00 |
| Coding index | 63.9 | 62.4 |
| Independent measurements behind it | 6 | 5 |
| Cost per solved task | $0.42 | $1.32 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.6 Terra better than Qwen3.8 Max (0902) for coding?
GPT-5.6 Terra and Qwen3.8 Max (0902) are level for coding, 63.9 against 62.4 on the Vibe Coding Index, so price and context decide. Pick Qwen3.8 Max (0902) for output tokens 2.0x cheaper ($6.00 vs $12.00 per million) and work measured by LMArena WebDev or LiveCodeBench (Vals AI), where it leads. Pick GPT-5.6 Terra for a lower cost per solved task ($0.42 against $1.32) and a larger context window (1.05M against 1M tokens).
Which is cheaper, GPT-5.6 Terra or Qwen3.8 Max (0902)?
GPT-5.6 Terra costs $2.00 per million input tokens and $12.00 per million output tokens; Qwen3.8 Max (0902) costs $2.00 and $6.00. Per solved SWE-bench Verified task, GPT-5.6 Terra costs $0.42 and Qwen3.8 Max (0902) $1.32.
Which has the bigger context window, GPT-5.6 Terra or Qwen3.8 Max (0902)?
GPT-5.6 Terra, with 1.05M tokens against 1M.
How many benchmarks is the GPT-5.6 Terra vs Qwen3.8 Max (0902) comparison based on?
107 benchmarks have a result for both models as of October 22, 2026. GPT-5.6 Terra's coding index rests on 6 independent benchmarks and Qwen3.8 Max (0902)'s on 5. Where only one model has a result, the page says so instead of naming a winner.