Claude Haiku 4.5 vs Qwen3.8 Max (0902)
Updated October 21, 2026. 77 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Qwen3.8 Max (0902) for coding: it scores 63.5 on the Vibe Coding Index against 49.8 for Claude Haiku 4.5, though on 2 benchmarks against 3.
Pick Claude Haiku 4.5 when output tokens cost 1.2x less, or when a solved task costs $0.56 against $1.32.
- Shared results
- 77
- Claude Haiku 4.5 only
- 17
- Qwen3.8 Max (0902) only
- 36
- 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
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) scores 63.5 against 49.8.
3 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 45.4 against 16.9.
Same board, both measured
Agentic work
AA Agentic Index
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 56.0 against 8.0.
Same board, both measured
Reasoning
GPQA Diamond
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 93.7% against 72.2%.
Same board, both measured
Math
AIME
No call
Only Claude Haiku 4.5 has a AIME result, 82.7%.
Directional only
Knowledge
MMLU Pro
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 88.6% against 78.7%.
Same board, both measured
Web design
Design Arena: websites
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) leads, 1275 against 1122.
Same board, both measured
Long documents
Listed context window
Qwen3.8 Max (0902)
Qwen3.8 Max (0902) takes 1M tokens against 200K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Claude Haiku 4.5
Claude Haiku 4.5 $0.0035, Qwen3.8 Max (0902) $0.0050 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
Claude Haiku 4.5
Claude Haiku 4.5 $0.56, Qwen3.8 Max (0902) $1.32.
Published attempt cost and pass rate
At a glance
- Coding index
- Claude Haiku 4.5 49.8 from 3 benchmarks, Qwen3.8 Max (0902) 63.5 from 2 benchmarks.
- Widest category
- Agentic work, 29.4 points apart on average over 8 shared benchmarks.
- Widest single gap
- Vibe Code Bench (Vals AI): Claude Haiku 4.5 11.4% against Qwen3.8 Max (0902) 64.7%, 53.3 points apart on a 0 to 100 scale.
- Output price
- Claude Haiku 4.5 $5.00 against Qwen3.8 Max (0902) $6.00 per million output tokens, 1.2x apart.
- Context window
- Claude Haiku 4.5 takes 200K 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 | Claude Haiku 4.5 | Bars | Qwen3.8 Max (0902) | Ties |
|---|---|---|---|---|
| Coding | 0 | 12 | 1 tie | |
| Intelligence | 0 | 2 | ||
| Agentic work | 1 | 7 | ||
| Reasoning | 0 | 8 | ||
| Math | 0 | 3 | ||
| Knowledge | 0 | 11 | ||
| Web design | 0 | 8 | ||
| General | 1 | 23 | ||
| All shared boards | 2 | 74 | 1 tie |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index012
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| SWE-bench Verified | 66.6% | Not measured | Only Claude Haiku 4.5 measured |
| SWE-bench Multilingual | 64.7% | Not measured | Only Claude Haiku 4.5 measured |
| LiveBench coding | 48.1% | 60.3% | Qwen3.8 Max (0902) |
| LMArena WebDev | 1330 | 1674 | Qwen3.8 Max (0902) |
| AA Coding Index | 43.9 | 76.2 | Qwen3.8 Max (0902) |
| LiveCodeBench (Vals AI) | 41.2% | 87.8% | Qwen3.8 Max (0902) |
| SWE-bench Verified (Vals AI) | 66.6% | 85.6% | Qwen3.8 Max (0902) |
| Terminal-Bench 4 (Vals AI) | Not measured | 34.3% | Only Qwen3.8 Max (0902) measured |
| Vibe Code Bench (Vals AI) | 11.4% | 64.7% | Qwen3.8 Max (0902) |
| Code migration (Vals AI) | 7.5% | 24.0% | Qwen3.8 Max (0902) |
| ProgramBench (Vals AI) | 0.0% | 0.0% | Tie |
| IOI (Vals AI) | Not measured | 68.9% | Only Qwen3.8 Max (0902) measured |
| Terminal-Bench 2.1 (Vals AI) | 43.8% | 67.4% | Qwen3.8 Max (0902) |
| SWE-bench Multimodal (Vals AI) | Not measured | 42.5% | Only Qwen3.8 Max (0902) measured |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 12.8% | Only Qwen3.8 Max (0902) measured |
| SkillsBench (Vals AI) | Not measured | 42.0% | Only Qwen3.8 Max (0902) measured |
| CyberBench (Vals AI) | Not measured | 28.6% | Only Qwen3.8 Max (0902) measured |
| SciCode | Not measured | 53.2% | Only Qwen3.8 Max (0902) measured |
| Terminal-Bench | 35.5% | Not measured | Only Claude Haiku 4.5 measured |
| FrontierSWE | Not measured | 17.8% | Only Qwen3.8 Max (0902) measured |
| WeirdML | 45.4% | Not measured | Only Claude Haiku 4.5 measured |
| LMArena Text: coding | 1454 | 1502 | Qwen3.8 Max (0902) |
| LMArena Text: software and IT | 1439 | 1498 | Qwen3.8 Max (0902) |
| LMArena WebDev: React | 1326 | 1692 | Qwen3.8 Max (0902) |
| LMArena WebDev: HTML | 1318 | 1649 | Qwen3.8 Max (0902) |
| LMArena WebDev: from an image | Not measured | 1637 | Only Qwen3.8 Max (0902) measured |
| Neither model has a result on the other 12 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Intelligence Index | 16.9 | 45.4 | Qwen3.8 Max (0902) |
| Epoch Capabilities Index | 142.2 | 157.4 | Qwen3.8 Max (0902) |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index17
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AA Agentic Index | 8.0 | 56.0 | Qwen3.8 Max (0902) |
| Design Arena: agents building web apps | Not measured | 1301 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building full-stack apps | Not measured | 1305 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building mobile apps | Not measured | 1223 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building Android apps | Not measured | 1207 | Only Qwen3.8 Max (0902) measured |
| Design Arena: agents building Godot games | Not measured | 1234 | Only Qwen3.8 Max (0902) measured |
| Design Arena: PowerPoint slides | Not measured | 1257 | Only Qwen3.8 Max (0902) measured |
| Design Arena: spreadsheets | Not measured | 868 | Only Qwen3.8 Max (0902) measured |
| LiveBench research | 42.2% | 69.9% | Qwen3.8 Max (0902) |
| LiveBench browser | 40.0% | 32.0% | Claude Haiku 4.5 |
| LiveBench long agentic | 11.1% | 42.3% | Qwen3.8 Max (0902) |
| Finance Agent (Vals AI) | 31.0% | 50.6% | Qwen3.8 Max (0902) |
| Tax Agent Bench (Vals AI) | 5.9% | 32.1% | Qwen3.8 Max (0902) |
| Legal Research Bench (Vals AI) | 10.6% | 47.6% | Qwen3.8 Max (0902) |
| Excel Modeling Benchmark (Vals AI) | 22.7% | 60.1% | Qwen3.8 Max (0902) |
| Terminal-Bench-Science (Vals AI) | Not measured | 12.9% | Only Qwen3.8 Max (0902) measured |
| APEX-Agents | Not measured | 63.3% | Only Qwen3.8 Max (0902) measured |
| BALROG | 31.2% | Not measured | Only Claude Haiku 4.5 measured |
| DeepResearch Bench | 45.5% | Not measured | Only Claude Haiku 4.5 measured |
| Vending-Bench 2 | $459 | Not measured | Only Claude Haiku 4.5 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 index08
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| GPQA Diamond | 72.2% | 93.7% | Qwen3.8 Max (0902) |
| LiveBench reasoning | 61.7% | 88.2% | Qwen3.8 Max (0902) |
| MysteryMechanism (Vals AI) | Not measured | 23.9% | Only Qwen3.8 Max (0902) measured |
| MMMU Pro (Vals AI) | 46.1% | 88.0% | Qwen3.8 Max (0902) |
| GPQA Diamond (Artificial Analysis run) | Not measured | 92.8% | Only Qwen3.8 Max (0902) measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 43.1% | Only Qwen3.8 Max (0902) measured |
| GPQA Diamond (Epoch AI run) | 71.2% | 92.7% | Qwen3.8 Max (0902) |
| ARC-AGI | 47.7% | Not measured | Only Claude Haiku 4.5 measured |
| ARC-AGI-2 | 4.0% | Not measured | Only Claude Haiku 4.5 measured |
| Chess puzzles | 8.0% | 46.0% | Qwen3.8 Max (0902) |
| Mystery game puzzles | Not measured | 38.0% | Only Qwen3.8 Max (0902) measured |
| LMArena Text: hard prompts | 1421 | 1496 | Qwen3.8 Max (0902) |
| LMArena Text: expert | 1443 | 1507 | Qwen3.8 Max (0902) |
| LMArena Text: hard prompts in English | 1435 | 1503 | Qwen3.8 Max (0902) |
| LMArena Vision: diagrams | Not measured | 1323 | Only Qwen3.8 Max (0902) measured |
| LMArena Vision: homework | Not measured | 1336 | Only Qwen3.8 Max (0902) measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index03
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| AIME | 82.7% | Not measured | Only Claude Haiku 4.5 measured |
| MGSM (Vals AI) | 92.2% | Not measured | Only Claude Haiku 4.5 measured |
| OTIS Mock AIME | 66.7% | 100.0% | Qwen3.8 Max (0902) |
| FrontierMath, tiers 1 to 3 | Not measured | 74.7% | Only Qwen3.8 Max (0902) measured |
| FrontierMath, tier 4 | Not measured | 46.3% | Only Qwen3.8 Max (0902) measured |
| ProofBench | Not measured | 58.0% | Only Qwen3.8 Max (0902) measured |
| MATH level 5 | 96.4% | Not measured | Only Claude Haiku 4.5 measured |
| LMArena Text: math | 1398 | 1501 | Qwen3.8 Max (0902) |
| LMArena Text: mathematical work | 1430 | 1491 | Qwen3.8 Max (0902) |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index011
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| MMLU Pro | 78.7% | 88.6% | Qwen3.8 Max (0902) |
| LegalBench (Vals AI) | 81.2% | 83.6% | Qwen3.8 Max (0902) |
| CaseLaw (Vals AI) | 56.5% | Not measured | Only Claude Haiku 4.5 measured |
| CorpFin (Vals AI) | 60.6% | 65.8% | Qwen3.8 Max (0902) |
| TaxEval (Vals AI) | 67.5% | 75.5% | Qwen3.8 Max (0902) |
| MortgageTax (Vals AI) | 62.2% | 64.0% | Qwen3.8 Max (0902) |
| MedQA (Vals AI) | 79.6% | Not measured | Only Claude Haiku 4.5 measured |
| MedCode (Vals AI) | 32.7% | 40.7% | Qwen3.8 Max (0902) |
| SimpleQA Verified | 13.2% | 47.3% | Qwen3.8 Max (0902) |
| LMArena Text: science | 1406 | 1512 | Qwen3.8 Max (0902) |
| LMArena Text: medicine and healthcare | 1394 | 1503 | Qwen3.8 Max (0902) |
| LMArena Text: legal and government | 1400 | 1492 | Qwen3.8 Max (0902) |
| LMArena Text: business and finance | 1400 | 1480 | Qwen3.8 Max (0902) |
| Neither model has a result on the other 4 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| Design Arena: websites | 1122 | 1275 | Qwen3.8 Max (0902) |
| Design Arena: UI components | 1100 | 1340 | Qwen3.8 Max (0902) |
| Design Arena: game dev | 1097 | 1313 | Qwen3.8 Max (0902) |
| Design Arena: data viz | 1124 | 1281 | Qwen3.8 Max (0902) |
| Design Arena: 3D | 1072 | 1337 | Qwen3.8 Max (0902) |
| Design Arena: all code categories | 1115 | 1293 | Qwen3.8 Max (0902) |
| Design Arena: SVG | 1025 | Not measured | Only Claude Haiku 4.5 measured |
| Design Arena: ASCII art | 1139 | 1201 | Qwen3.8 Max (0902) |
| LiveBench design | 59.9% | 78.6% | Qwen3.8 Max (0902) |
GeneralLike-for-likePublic boards, shown as published, kept out of the index123
| Benchmark | Claude Haiku 4.5 | Qwen3.8 Max (0902) | Leader |
|---|---|---|---|
| LiveBench data analysis | 59.3% | 78.4% | Qwen3.8 Max (0902) |
| Public Benefits Bench (Vals AI) | 54.3% | 67.1% | Qwen3.8 Max (0902) |
| SAGE (Vals AI) | 31.8% | 51.2% | Qwen3.8 Max (0902) |
| MedScribe (Vals AI) | 85.2% | 85.0% | Claude Haiku 4.5 |
| Vals Multimodal Index | 42.9% | 65.4% | Qwen3.8 Max (0902) |
| Long context reasoning (AA-LCR) | Not measured | 80.3% | Only Qwen3.8 Max (0902) measured |
| DTBench | 73.6% | 92.0% | Qwen3.8 Max (0902) |
| LMCA | 30.9% | 46.2% | Qwen3.8 Max (0902) |
| Furniture assembly | Not measured | 20.0% | Only Qwen3.8 Max (0902) measured |
| ExploitBench | 13.7% | Not measured | Only Claude Haiku 4.5 measured |
| LMArena Text | 1397 | 1483 | Qwen3.8 Max (0902) |
| LMArena Text: instruction following | 1414 | 1481 | Qwen3.8 Max (0902) |
| LMArena Text: creative writing | 1371 | 1477 | Qwen3.8 Max (0902) |
| LMArena Vision | Not measured | 1313 | Only Qwen3.8 Max (0902) measured |
| LMArena Text: English | 1414 | 1491 | Qwen3.8 Max (0902) |
| LMArena Text: non-English | 1377 | 1473 | Qwen3.8 Max (0902) |
| LMArena Text: Chinese | 1418 | 1539 | Qwen3.8 Max (0902) |
| LMArena Text: German | 1374 | 1478 | Qwen3.8 Max (0902) |
| LMArena Text: Japanese | 1338 | 1473 | Qwen3.8 Max (0902) |
| LMArena Text: Korean | 1346 | 1460 | Qwen3.8 Max (0902) |
| LMArena Text: French | 1411 | 1498 | Qwen3.8 Max (0902) |
| LMArena Text: Spanish | 1421 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: Russian | 1381 | 1482 | Qwen3.8 Max (0902) |
| LMArena Text: Polish | 1343 | 1471 | Qwen3.8 Max (0902) |
| LMArena Text: multi-turn | 1409 | 1489 | Qwen3.8 Max (0902) |
| LMArena Text: longer queries | 1427 | 1488 | Qwen3.8 Max (0902) |
| LMArena Text: writing and language | 1386 | 1477 | Qwen3.8 Max (0902) |
| LMArena Text: entertainment and media | 1369 | 1464 | Qwen3.8 Max (0902) |
| LMArena Vision: English | Not measured | 1311 | Only Qwen3.8 Max (0902) measured |
| LMArena Vision: Chinese | Not measured | 1358 | Only Qwen3.8 Max (0902) measured |
| LMArena Vision: reading text | Not measured | 1322 | Only Qwen3.8 Max (0902) measured |
| LMArena Vision: humor | Not measured | 1304 | Only Qwen3.8 Max (0902) measured |
| LMArena Vision: creative writing | Not measured | 1312 | Only Qwen3.8 Max (0902) measured |
| LMArena Document | 1420 | Not measured | Only Claude Haiku 4.5 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 Haiku 4.5 | 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
Claude Haiku 4.5 costs less
Repository review
50K input + 3K output tokens
Claude Haiku 4.5 costs less
Agent loop
200K input + 10K output tokens
Claude Haiku 4.5 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 Haiku 4.5 | Qwen3.8 Max (0902) |
|---|---|---|
| Maker | Anthropic | Qwen |
| Weights | No public weights listed | No public weights listed |
| Released | October 15, 2025 | September 3, 2026 |
| Context window | 200K | 1M |
| Input $/Mtok | $1.00 | $2.00 |
| Output $/Mtok | $5.00 | $6.00 |
| Coding index | 49.8 | 63.5 |
| Independent measurements behind it | 3 | 2 |
| Cost per solved task | $0.56 | $1.32 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Haiku 4.5 better than Qwen3.8 Max (0902) for coding?
Pick Qwen3.8 Max (0902) for coding: it scores 63.5 on the Vibe Coding Index against 49.8 for Claude Haiku 4.5, though on 2 benchmarks against 3. Pick Claude Haiku 4.5 when output tokens cost 1.2x less, or when a solved task costs $0.56 against $1.32.
Which is cheaper, Claude Haiku 4.5 or Qwen3.8 Max (0902)?
Claude Haiku 4.5 costs $1.00 per million input tokens and $5.00 per million output tokens; Qwen3.8 Max (0902) costs $2.00 and $6.00. Per solved SWE-bench Verified task, Claude Haiku 4.5 costs $0.56 and Qwen3.8 Max (0902) $1.32.
Which has the bigger context window, Claude Haiku 4.5 or Qwen3.8 Max (0902)?
Qwen3.8 Max (0902), with 1M tokens against 200K.
How many benchmarks is the Claude Haiku 4.5 vs Qwen3.8 Max (0902) comparison based on?
77 benchmarks have a result for both models as of October 21, 2026. Claude Haiku 4.5's coding index rests on 3 independent benchmarks and Qwen3.8 Max (0902)'s on 2. Where only one model has a result, the page says so instead of naming a winner.