Qwen3.8 27B vs Hy4
Qwen3.8 27B leads Hy4 on coding, 74.1 to 65.2 on the Vibe Coding Index, and Hy4 costs 2.0x less per output token ($1.2505 vs $2.55 per million). Updated November 5, 2026; 3 benchmarks have a result for both.
Qwen3.8 27B or Hy4: which to pick
Pick Qwen3.8 27B for coding: it scores 74.1 on the Vibe Coding Index against 65.2 for Hy4.
Pick Hy4 for output tokens 2.0x cheaper ($1.2505 vs $2.55 per million) and a larger context window (1.05M against 1.02M tokens).
- Shared results
- 3
- Qwen3.8 27B only
- 82
- Hy4 only
- 3
- Like-for-like categories
- 2 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
Qwen3.8 27B
Qwen3.8 27B scores 74.1 against 65.2.
4 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
No call
Only Qwen3.8 27B has a AA Intelligence Index result, 20.1.
Directional only
Agentic work
AA Agentic Index
No call
Only Qwen3.8 27B has a AA Agentic Index result, 45.8.
Directional only
Reasoning
GPQA Diamond
No call
Only Qwen3.8 27B has a GPQA Diamond result, 88.9%.
Directional only
Knowledge
MMLU Pro
No call
Only Qwen3.8 27B has a MMLU Pro result, 84.3%.
Directional only
General
Vals Index
No call
Only Hy4 has a Vals Index result, 59.4%.
Directional only
Long documents
Listed context window
Hy4
Hy4 takes 1.05M tokens against 1.02M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy4
Qwen3.8 27B $0.0017, Hy4 $0.0010 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Qwen3.8 27B has the number, $1.26.
Directional only
At a glance
- Coding index
- Qwen3.8 27B 74.1 from 4 benchmarks, Hy4 65.2 from 2 benchmarks.
- Widest category
- Coding, 16.2 points apart on average over 1 shared benchmark.
- Widest single gap
- Vibe Code Bench (Vals AI): Qwen3.8 27B 64.8% against Hy4 81.0%, 16.2 points apart on a 0 to 100 scale.
- Output price
- Qwen3.8 27B $2.55 against Hy4 $1.2505 per million output tokens, 2.0x apart.
- Context window
- Qwen3.8 27B takes 1.02M tokens, Hy4 1.05M.
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 | Qwen3.8 27B | Bars | Hy4 | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Agentic work | 0 | 2 | ||
| All shared boards | 0 | 3 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index01
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| LiveBench coding | 86.8% | Not measured | Only Qwen3.8 27B measured |
| LMArena WebDev | 1590 | Not measured | Only Qwen3.8 27B measured |
| AA Coding Index | 68.1 | Not measured | Only Qwen3.8 27B measured |
| LiveCodeBench (Vals AI) | 84.0% | Not measured | Only Qwen3.8 27B measured |
| SWE-bench Verified (Vals AI) | 86.0% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 39.4% | Only Hy4 measured |
| Vibe Code Bench (Vals AI) | 64.8% | 81.0% | Hy4 |
| Code migration (Vals AI) | 14.2% | Not measured | Only Qwen3.8 27B measured |
| ProgramBench (Vals AI) | 0.0% | Not measured | Only Qwen3.8 27B measured |
| IOI (Vals AI) | 39.1% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench 2.1 (Vals AI) | 58.4% | Not measured | Only Qwen3.8 27B measured |
| SkillsBench (Vals AI) | 38.1% | Not measured | Only Qwen3.8 27B measured |
| CyberBench (Vals AI) | 67.8% | Not measured | Only Qwen3.8 27B measured |
| SciCode | 46.6% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: coding | 1481 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: software and IT | 1477 | Not measured | Only Qwen3.8 27B measured |
| LMArena WebDev: React | 1598 | Not measured | Only Qwen3.8 27B measured |
| LMArena WebDev: HTML | 1565 | Not measured | Only Qwen3.8 27B measured |
| LMArena WebDev: from an image | 1577 | Not measured | Only Qwen3.8 27B measured |
| GameBench 2 | 43.6 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 30 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.1 | Not measured | Only Qwen3.8 27B measured |
| Epoch Capabilities Index | 152.7 | Not measured | Only Qwen3.8 27B measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| AA Agentic Index | 45.8 | Not measured | Only Qwen3.8 27B measured |
| LiveBench research | 59.4% | Not measured | Only Qwen3.8 27B measured |
| LiveBench browser | 24.2% | Not measured | Only Qwen3.8 27B measured |
| LiveBench long agentic | 20.6% | Not measured | Only Qwen3.8 27B measured |
| Finance Agent (Vals AI) | 48.5% | Not measured | Only Qwen3.8 27B measured |
| Tax Agent Bench (Vals AI) | 30.9% | 34.4% | Hy4 |
| Legal Research Bench (Vals AI) | 36.1% | 41.4% | Hy4 |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 12.5% | Only Hy4 measured |
| Excel Modeling Benchmark (Vals AI) | 59.7% | Not measured | Only Qwen3.8 27B measured |
| Terminal-Bench-Science (Vals AI) | 0.0% | Not measured | Only Qwen3.8 27B measured |
| APEX-Agents | 47.5% | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 33 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index12 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| GPQA Diamond | 88.9% | Not measured | Only Qwen3.8 27B measured |
| LiveBench reasoning | 80.0% | Not measured | Only Qwen3.8 27B measured |
| MMMU Pro (Vals AI) | 83.9% | Not measured | Only Qwen3.8 27B measured |
| GPQA Diamond (Artificial Analysis run) | 90.5% | Not measured | Only Qwen3.8 27B measured |
| Humanity's Last Exam (Artificial Analysis run) | 33.9% | Not measured | Only Qwen3.8 27B measured |
| ARC-AGI | 87.5% | Not measured | Only Qwen3.8 27B measured |
| ARC-AGI-2 | 42.4% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: hard prompts | 1458 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: expert | 1482 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: hard prompts in English | 1460 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: diagrams | 1285 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: homework | 1295 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 12 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| ProofBench | 16.0% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: math | 1456 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: mathematical work | 1460 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 11 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| MMLU Pro | 84.3% | Not measured | Only Qwen3.8 27B measured |
| LegalBench (Vals AI) | 82.4% | Not measured | Only Qwen3.8 27B measured |
| TaxEval (Vals AI) | 70.8% | Not measured | Only Qwen3.8 27B measured |
| MortgageTax (Vals AI) | 64.9% | Not measured | Only Qwen3.8 27B measured |
| MedCode (Vals AI) | 28.7% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: science | 1468 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: medicine and healthcare | 1442 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: legal and government | 1447 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: business and finance | 1439 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 11 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index1 board, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| LiveBench design | 75.9% | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 8 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index30 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | Leader |
|---|---|---|---|
| Vals Index | Not measured | 59.4% | Only Hy4 measured |
| LiveBench data analysis | 76.6% | Not measured | Only Qwen3.8 27B measured |
| MedScribe (Vals AI) | 83.8% | Not measured | Only Qwen3.8 27B measured |
| Long context reasoning (AA-LCR) | 82.0% | Not measured | Only Qwen3.8 27B measured |
| DTBench | 88.0% | Not measured | Only Qwen3.8 27B measured |
| LMCA | 41.4% | Not measured | Only Qwen3.8 27B measured |
| Surface Evolver Bench | 45.0% | Not measured | Only Qwen3.8 27B measured |
| LMArena Text | 1438 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: instruction following | 1435 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: creative writing | 1379 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision | 1269 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: English | 1447 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: non-English | 1426 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Chinese | 1501 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: German | 1444 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Japanese | 1396 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Korean | 1380 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: French | 1465 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Spanish | 1446 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Russian | 1411 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: Polish | 1446 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: multi-turn | 1438 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: longer queries | 1448 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: writing and language | 1401 | Not measured | Only Qwen3.8 27B measured |
| LMArena Text: entertainment and media | 1390 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: English | 1267 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: Chinese | 1326 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: reading text | 1281 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: humor | 1235 | Not measured | Only Qwen3.8 27B measured |
| LMArena Vision: creative writing | 1237 | Not measured | Only Qwen3.8 27B measured |
| Neither model has a result on the other 18 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Qwen3.8 27B | Hy4 | 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
Hy4 costs less
Repository review
50K input + 3K output tokens
Hy4 costs less
Agent loop
200K input + 10K output tokens
Hy4 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 | Qwen3.8 27B | Tencent Hy4 |
|---|---|---|
| Maker | Qwen | Tencent |
| Weights | Open weights | Open weights |
| Released | August 14, 2026 | November 4, 2026 |
| Context window | 1.02M | 1.05M |
| Input $/Mtok | $0.425 | $0.417 |
| Output $/Mtok | $2.55 | $1.2505 |
| Coding index | 74.1 | 65.2 |
| Independent measurements behind it | 4 | 2 |
| Cost per solved task | $1.26 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Qwen3.8 27B better than Hy4 for coding?
Pick Qwen3.8 27B for coding: it scores 74.1 on the Vibe Coding Index against 65.2 for Hy4. Pick Hy4 for output tokens 2.0x cheaper ($1.2505 vs $2.55 per million) and a larger context window (1.05M against 1.02M tokens).
Which is cheaper, Qwen3.8 27B or Hy4?
Qwen3.8 27B costs $0.425 per million input tokens and $2.55 per million output tokens; Hy4 costs $0.417 and $1.2505. Cost per solved task is known for Qwen3.8 27B only.
Which has the bigger context window, Qwen3.8 27B or Hy4?
Hy4, with 1.05M tokens against 1.02M.
How many benchmarks is the Qwen3.8 27B vs Hy4 comparison based on?
3 benchmarks have a result for both models as of November 5, 2026. Qwen3.8 27B's coding index rests on 4 independent benchmarks and Hy4's on 2. Where only one model has a result, the page says so instead of naming a winner.