gpt-oss-120b vs Hy3 preview
Updated October 21, 2026. 33 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick gpt-oss-120b for coding: it scores 34.5 on the Vibe Coding Index against 22.4 for Hy3 preview.
Pick Hy3 preview when the context window is larger, 262K against 131K tokens, or when what matters is AA Coding Index or Terminal-Bench Hard (Artificial Analysis run), where it leads.
- Shared results
- 33
- gpt-oss-120b only
- 45
- Hy3 preview only
- 3
- Like-for-like categories
- 7 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
gpt-oss-120b
gpt-oss-120b scores 34.5 against 22.4.
2 and 1 benchmarks behind the scores
Intelligence
AA Intelligence Index
Hy3 preview
Hy3 preview leads, 25.3 against 11.6.
Same board, both measured
Agentic work
AA Agentic Index
Hy3 preview
Hy3 preview leads, 24.1 against 3.7.
Same board, both measured
Reasoning
GPQA Diamond
No call
Only gpt-oss-120b has a GPQA Diamond result, 78.5%.
Directional only
Math
AIME
No call
Only gpt-oss-120b has a AIME result, 92.6%.
Directional only
Knowledge
MMLU Pro
No call
Only gpt-oss-120b has a MMLU Pro result, 79.2%.
Directional only
Web design
Design Arena: websites
No call
Only gpt-oss-120b has a Design Arena: websites result, 968.
Directional only
Long documents
Listed context window
Hy3 preview
Hy3 preview takes 262K tokens against 131K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
gpt-oss-120b
gpt-oss-120b $0.0004, Hy3 preview $0.0005 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only gpt-oss-120b has the number, $0.51.
Directional only
At a glance
- Coding index
- gpt-oss-120b 34.5 from 2 benchmarks, Hy3 preview 22.4 from 1 benchmark.
- Widest category
- Agentic work, 23.7 points apart on average over 2 shared benchmarks.
- Widest single gap
- AA Coding Index: gpt-oss-120b 30.4 against Hy3 preview 58.8, 28.4 points apart on a 0 to 100 scale.
- Output price
- Both cost $0.60 per million output tokens.
- Context window
- gpt-oss-120b takes 131K tokens, Hy3 preview 262K.
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-oss-120b | Bars | Hy3 preview | Ties |
|---|---|---|---|---|
| Coding | 0 | 4 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 2 | ||
| Reasoning | 0 | 5 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 4 | ||
| General | 1 | 14 | ||
| All shared boards | 1 | 32 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index04
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| Aider polyglot | 41.8% | Not measured | Only gpt-oss-120b measured |
| SWE-bench Verified | 26.0% | Not measured | Only gpt-oss-120b measured |
| LMArena WebDev | Not measured | 1357 | Only Hy3 preview measured |
| AA Coding Index | 30.4 | 58.8 | Hy3 preview |
| LiveCodeBench (Vals AI) | 83.2% | Not measured | Only gpt-oss-120b measured |
| SWE-bench Verified (Vals AI) | 33.6% | Not measured | Only gpt-oss-120b measured |
| LiveCodeBench (Artificial Analysis run) | 87.8% | Not measured | Only gpt-oss-120b measured |
| SciCode | 34.0% | Not measured | Only gpt-oss-120b measured |
| Terminal-Bench Hard (Artificial Analysis run) | 23.5% | 34.1% | Hy3 preview |
| Terminal-Bench | 18.7% | Not measured | Only gpt-oss-120b measured |
| WeirdML | 48.2% | Not measured | Only gpt-oss-120b measured |
| ALE-Bench | 575.6 | Not measured | Only gpt-oss-120b measured |
| LMArena Text: coding | 1379 | 1421 | Hy3 preview |
| LMArena Text: software and IT | 1387 | 1423 | Hy3 preview |
| LMArena WebDev: React | Not measured | 1349 | Only Hy3 preview measured |
| LMArena WebDev: HTML | Not measured | 1356 | Only Hy3 preview measured |
| Neither model has a result on the other 22 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 11.6 | 25.3 | Hy3 preview |
| Epoch Capabilities Index | 139.8 | Not measured | Only gpt-oss-120b measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| AA Agentic Index | 3.7 | 24.1 | Hy3 preview |
| τ²-Bench Telecom | 65.8% | 92.7% | Hy3 preview |
| APEX-Agents | 4.4% | Not measured | Only gpt-oss-120b measured |
| METR time horizons | 56.6% | Not measured | Only gpt-oss-120b measured |
| Vending-Bench 2 | -$22 | Not measured | Only gpt-oss-120b measured |
| Neither model has a result on the other 34 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| GPQA Diamond | 78.5% | Not measured | Only gpt-oss-120b measured |
| GPQA Diamond (Artificial Analysis run) | 78.2% | 86.7% | Hy3 preview |
| Humanity's Last Exam (Artificial Analysis run) | 19.6% | 27.8% | Hy3 preview |
| GPQA Diamond (Epoch AI run) | 75.8% | Not measured | Only gpt-oss-120b measured |
| SimpleBench | 22.1% | Not measured | Only gpt-oss-120b measured |
| Mystery game puzzles | 2.0% | Not measured | Only gpt-oss-120b measured |
| LMArena Text: hard prompts | 1364 | 1413 | Hy3 preview |
| LMArena Text: expert | 1356 | 1422 | Hy3 preview |
| LMArena Text: hard prompts in English | 1368 | 1416 | Hy3 preview |
| Neither model has a result on the other 12 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| AIME | 92.6% | Not measured | Only gpt-oss-120b measured |
| MATH 500 (Vals AI) | 94.8% | Not measured | Only gpt-oss-120b measured |
| MGSM (Vals AI) | 92.0% | Not measured | Only gpt-oss-120b measured |
| AA Math Index | 93.4 | Not measured | Only gpt-oss-120b measured |
| AIME 2025 (Artificial Analysis run) | 93.4% | Not measured | Only gpt-oss-120b measured |
| OTIS Mock AIME | 88.9% | Not measured | Only gpt-oss-120b measured |
| LMArena Text: math | 1389 | 1424 | Hy3 preview |
| LMArena Text: mathematical work | 1386 | 1442 | Hy3 preview |
| Neither model has a result on the other 5 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| MMLU Pro | 79.2% | Not measured | Only gpt-oss-120b measured |
| LegalBench (Vals AI) | 75.9% | Not measured | Only gpt-oss-120b measured |
| CaseLaw (Vals AI) | 48.8% | Not measured | Only gpt-oss-120b measured |
| CorpFin (Vals AI) | 58.2% | Not measured | Only gpt-oss-120b measured |
| TaxEval (Vals AI) | 71.6% | Not measured | Only gpt-oss-120b measured |
| MedQA (Vals AI) | 91.4% | Not measured | Only gpt-oss-120b measured |
| MMLU Pro (Artificial Analysis run) | 80.8% | Not measured | Only gpt-oss-120b measured |
| LMArena Text: science | 1382 | 1416 | Hy3 preview |
| LMArena Text: medicine and healthcare | 1388 | 1405 | Hy3 preview |
| LMArena Text: legal and government | 1355 | 1430 | Hy3 preview |
| LMArena Text: business and finance | 1359 | 1397 | Hy3 preview |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| Design Arena: websites | 968 | Not measured | Only gpt-oss-120b measured |
| Design Arena: UI components | 927 | Not measured | Only gpt-oss-120b measured |
| Design Arena: game dev | 992 | Not measured | Only gpt-oss-120b measured |
| Design Arena: data viz | 987 | Not measured | Only gpt-oss-120b measured |
| Design Arena: 3D | 901 | Not measured | Only gpt-oss-120b measured |
| Design Arena: all code categories | 963 | Not measured | Only gpt-oss-120b measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index114
| Benchmark | gpt-oss-120b | Hy3 preview | Leader |
|---|---|---|---|
| IFBench | 69.0% | 63.1% | gpt-oss-120b |
| Long context reasoning (AA-LCR) | 52.0% | 64.7% | Hy3 preview |
| DTBench | 76.3% | Not measured | Only gpt-oss-120b measured |
| LMCA | 22.1% | Not measured | Only gpt-oss-120b measured |
| Fiction.LiveBench | 44.4% | Not measured | Only gpt-oss-120b measured |
| Creative writing (Lech Mazur) | 77.1% | Not measured | Only gpt-oss-120b measured |
| Surface Evolver Bench | 25.0% | Not measured | Only gpt-oss-120b measured |
| LMArena Text | 1366 | 1401 | Hy3 preview |
| LMArena Text: instruction following | 1318 | 1383 | Hy3 preview |
| LMArena Text: creative writing | 1274 | 1339 | Hy3 preview |
| LMArena Text: English | 1375 | 1409 | Hy3 preview |
| LMArena Text: non-English | 1351 | 1390 | Hy3 preview |
| LMArena Text: Chinese | 1385 | 1435 | Hy3 preview |
| LMArena Text: German | 1353 | Not measured | Only gpt-oss-120b measured |
| LMArena Text: Japanese | 1331 | Not measured | Only gpt-oss-120b measured |
| LMArena Text: Korean | 1282 | Not measured | Only gpt-oss-120b measured |
| LMArena Text: French | 1370 | 1425 | Hy3 preview |
| LMArena Text: Spanish | 1389 | 1399 | Hy3 preview |
| LMArena Text: Russian | 1343 | 1399 | Hy3 preview |
| LMArena Text: Polish | 1353 | Not measured | Only gpt-oss-120b measured |
| LMArena Text: multi-turn | 1340 | 1411 | Hy3 preview |
| LMArena Text: longer queries | 1319 | 1402 | Hy3 preview |
| LMArena Text: writing and language | 1306 | 1370 | Hy3 preview |
| LMArena Text: entertainment and media | 1292 | 1352 | Hy3 preview |
| Neither model has a result on the other 22 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | gpt-oss-120b | Hy3 preview | 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-120b costs less
Repository review
50K input + 3K output tokens
gpt-oss-120b costs less
Agent loop
200K input + 10K output tokens
gpt-oss-120b 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-oss-120b | Tencent Hy3 preview |
|---|---|---|
| Maker | OpenAI | Tencent |
| Weights | Open weights | Open weights |
| Released | August 5, 2025 | April 22, 2026 |
| Context window | 131K | 262K |
| Input $/Mtok | $0.15 | $0.18 |
| Output $/Mtok | $0.60 | $0.60 |
| Coding index | 34.5 | 22.4 |
| Independent measurements behind it | 2 | 1 |
| Cost per solved task | $0.51 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is gpt-oss-120b better than Hy3 preview for coding?
Pick gpt-oss-120b for coding: it scores 34.5 on the Vibe Coding Index against 22.4 for Hy3 preview. Pick Hy3 preview when the context window is larger, 262K against 131K tokens, or when what matters is AA Coding Index or Terminal-Bench Hard (Artificial Analysis run), where it leads.
Which is cheaper, gpt-oss-120b or Hy3 preview?
gpt-oss-120b costs $0.15 per million input tokens and $0.60 per million output tokens; Hy3 preview costs $0.18 and $0.60. Cost per solved task is known for gpt-oss-120b only.
Which has the bigger context window, gpt-oss-120b or Hy3 preview?
Hy3 preview, with 262K tokens against 131K.
How many benchmarks is the gpt-oss-120b vs Hy3 preview comparison based on?
33 benchmarks have a result for both models as of October 21, 2026. gpt-oss-120b's coding index rests on 2 independent benchmarks and Hy3 preview's on 1. Where only one model has a result, the page says so instead of naming a winner.