Qwen3.5 397B A17B vs Hy3
Hy3 costs 4.4x less per output token than Qwen3.5 397B A17B ($0.528 vs $2.34 per million). Updated October 22, 2026; 43 benchmarks have a result for both.
Qwen3.5 397B A17B or Hy3: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick Qwen3.5 397B A17B for work measured by LMArena WebDev or SciCode, where it leads.
- Shared results
- 43
- Qwen3.5 397B A17B only
- 23
- Hy3 only
- 1
- Like-for-like categories
- 6 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
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
No call
Only Qwen3.5 397B A17B has a AA Intelligence Index result, 21.4.
Directional only
Web design
Design Arena: websites
Hy3
Hy3 leads, 1230 against 1190.
Same board, both measured
Long documents
Listed context window
No call
Both take 262K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy3
Qwen3.5 397B A17B $0.0016, Hy3 $0.0004 per turn.
List prices
At a glance
- Coding index
- Qwen3.5 397B A17B not benchmarked, Hy3 not benchmarked.
- Widest category
- Coding, 8.6 points apart on average over 7 shared benchmarks.
- Widest single gap
- LMArena WebDev: React: Qwen3.5 397B A17B 1388 against Hy3 1511, 20.5 points apart on a 0 to 100 scale.
- Output price
- Qwen3.5 397B A17B $2.34 against Hy3 $0.528 per million output tokens, 4.4x apart.
- Context window
- Both take 262K tokens.
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 | Qwen3.5 397B A17B | Bars | Hy3 | Ties |
|---|---|---|---|---|
| Coding | 1 | 6 | ||
| Reasoning | 2 | 3 | ||
| Math | 0 | 2 | ||
| Knowledge | 3 | 1 | ||
| Web design | 1 | 6 | ||
| General | 7 | 11 | ||
| All shared boards | 14 | 29 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index16
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| LMArena WebDev | 1400 | 1507 | Hy3 |
| SciCode | 44.8% | 48.6% | Hy3 |
| Terminal-Bench Hard (Artificial Analysis run) | 40.9% | Not measured | Only Qwen3.5 397B A17B measured |
| ALE-Bench | 621.9 | 838.6 | Hy3 |
| LMArena Text: coding | 1464 | 1466 | Hy3 |
| LMArena Text: software and IT | 1463 | 1463 | Qwen3.5 397B A17B |
| LMArena WebDev: React | 1388 | 1511 | Hy3 |
| LMArena WebDev: HTML | 1410 | 1487 | Hy3 |
| Neither model has a result on the other 35 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 21.4 | Not measured | Only Qwen3.5 397B A17B measured |
| Epoch Capabilities Index | 146.8 | Not measured | Only Qwen3.5 397B A17B measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| τ²-Bench Telecom | 95.6% | Not measured | Only Qwen3.5 397B A17B measured |
| APEX-Agents | 24.9% | Not measured | Only Qwen3.5 397B A17B measured |
| Neither model has a result on the other 40 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index23
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 89.3% | 89.7% | Hy3 |
| Humanity's Last Exam (Artificial Analysis run) | 29.0% | 33.5% | Hy3 |
| GPQA Diamond (Epoch AI run) | 86.4% | Not measured | Only Qwen3.5 397B A17B measured |
| Mystery game puzzles | 18.0% | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Text: hard prompts | 1448 | 1447 | Qwen3.5 397B A17B |
| LMArena Text: expert | 1462 | 1461 | Qwen3.5 397B A17B |
| LMArena Text: hard prompts in English | 1450 | 1455 | Hy3 |
| LMArena Vision: diagrams | 1277 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: homework | 1292 | Not measured | Only Qwen3.5 397B A17B measured |
| Neither model has a result on the other 14 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| OTIS Mock AIME | 88.9% | Not measured | Only Qwen3.5 397B A17B measured |
| FrontierMath, tiers 1 to 3 | 31.2% | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Text: math | 1453 | 1473 | Hy3 |
| LMArena Text: mathematical work | 1454 | 1482 | Hy3 |
| MathArena | 19.2% | Not measured | Only Qwen3.5 397B A17B measured |
| Neither model has a result on the other 9 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index31
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| LMArena Text: science | 1457 | 1457 | Hy3 |
| LMArena Text: medicine and healthcare | 1455 | 1452 | Qwen3.5 397B A17B |
| LMArena Text: legal and government | 1441 | 1422 | Qwen3.5 397B A17B |
| LMArena Text: business and finance | 1435 | 1424 | Qwen3.5 397B A17B |
| Neither model has a result on the other 16 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index16
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| Design Arena: websites | 1190 | 1230 | Hy3 |
| Design Arena: UI components | 1161 | 1204 | Hy3 |
| Design Arena: game dev | 1140 | 1208 | Hy3 |
| Design Arena: data viz | 1175 | 1165 | Qwen3.5 397B A17B |
| Design Arena: 3D | 1160 | 1232 | Hy3 |
| Design Arena: all code categories | 1179 | 1227 | Hy3 |
| Design Arena: SVG | 1139 | 1201 | Hy3 |
| Design Arena: ASCII art | Not measured | 1152 | Only Hy3 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index711
| Benchmark | Qwen3.5 397B A17B | Hy3 | Leader |
|---|---|---|---|
| IFBench | 78.8% | Not measured | Only Qwen3.5 397B A17B measured |
| Long context reasoning (AA-LCR) | 77.3% | 79.0% | Hy3 |
| DTBench | 87.5% | Not measured | Only Qwen3.5 397B A17B measured |
| LMCA | 37.9% | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Text | 1437 | 1438 | Hy3 |
| LMArena Text: instruction following | 1424 | 1426 | Hy3 |
| LMArena Text: creative writing | 1402 | 1403 | Hy3 |
| LMArena Vision | 1262 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Text: English | 1443 | 1448 | Hy3 |
| LMArena Text: non-English | 1430 | 1426 | Qwen3.5 397B A17B |
| LMArena Text: Chinese | 1498 | 1485 | Qwen3.5 397B A17B |
| LMArena Text: German | 1445 | 1449 | Hy3 |
| LMArena Text: Japanese | 1429 | 1397 | Qwen3.5 397B A17B |
| LMArena Text: Korean | 1384 | 1396 | Hy3 |
| LMArena Text: French | 1462 | 1456 | Qwen3.5 397B A17B |
| LMArena Text: Spanish | 1441 | 1457 | Hy3 |
| LMArena Text: Russian | 1430 | 1432 | Hy3 |
| LMArena Text: Polish | 1447 | 1447 | Qwen3.5 397B A17B |
| LMArena Text: multi-turn | 1446 | 1436 | Qwen3.5 397B A17B |
| LMArena Text: longer queries | 1441 | 1442 | Hy3 |
| LMArena Text: writing and language | 1415 | 1414 | Qwen3.5 397B A17B |
| LMArena Text: entertainment and media | 1399 | 1400 | Hy3 |
| LMArena Vision: English | 1268 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: Chinese | 1314 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: reading text | 1271 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: humor | 1262 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: captioning | 1230 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: recognizing things | 1265 | Not measured | Only Qwen3.5 397B A17B measured |
| LMArena Vision: creative writing | 1260 | Not measured | Only Qwen3.5 397B A17B measured |
| Neither model has a result on the other 19 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Qwen3.5 397B A17B | Hy3 | 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
Hy3 costs less
Repository review
50K input + 3K output tokens
Hy3 costs less
Agent loop
200K input + 10K output tokens
Hy3 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 | Qwen3.5 397B A17B | Tencent Hy3 |
|---|---|---|
| Maker | Qwen | Tencent |
| Weights | Open weights | Open weights |
| Released | February 16, 2026 | July 6, 2026 |
| Context window | 262K | 262K |
| Input $/Mtok | $0.39 | $0.132 |
| Output $/Mtok | $2.34 | $0.528 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Qwen3.5 397B A17B better than Hy3 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Qwen3.5 397B A17B for work measured by LMArena WebDev or SciCode, where it leads.
Which is cheaper, Qwen3.5 397B A17B or Hy3?
Qwen3.5 397B A17B costs $0.39 per million input tokens and $2.34 per million output tokens; Hy3 costs $0.132 and $0.528. Cost per solved task is known for neither of them.
Which has the bigger context window, Qwen3.5 397B A17B or Hy3?
Neither: both take 262K tokens.
How many benchmarks is the Qwen3.5 397B A17B vs Hy3 comparison based on?
43 benchmarks have a result for both models as of October 22, 2026. Qwen3.5 397B A17B's coding index rests on 1 independent benchmark and Hy3's on 1. Where only one model has a result, the page says so instead of naming a winner.