GPT-5.4 Nano vs Hy3
Hy3 costs 2.4x less per output token than GPT-5.4 Nano ($0.528 vs $1.25 per million). Updated October 22, 2026; 33 benchmarks have a result for both.
GPT-5.4 Nano or Hy3: which to pick
Only GPT-5.4 Nano has a coding score, 44.8 on the Vibe Coding Index; Hy3 has none, so the two cannot be ranked on coding.
Pick Hy3 for output tokens 2.4x cheaper ($0.528 vs $1.25 per million) and work measured by SciCode or LMArena Text: coding, where it leads.
- Shared results
- 33
- GPT-5.4 Nano only
- 52
- Hy3 only
- 11
- Like-for-like categories
- 5 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
Only GPT-5.4 Nano has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
No call
Only GPT-5.4 Nano has a AA Intelligence Index result, 20.7.
Directional only
Agentic work
AA Agentic Index
No call
Only GPT-5.4 Nano has a AA Agentic Index result, 16.0.
Directional only
Reasoning
GPQA Diamond
No call
Only GPT-5.4 Nano has a GPQA Diamond result, 77.5%.
Directional only
Math
AIME
No call
Only GPT-5.4 Nano has a AIME result, 88.8%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5.4 Nano has a MMLU Pro result, 77.2%.
Directional only
Web design
Design Arena: websites
No call
Only Hy3 has a Design Arena: websites result, 1230.
Directional only
Long documents
Listed context window
GPT-5.4 Nano
GPT-5.4 Nano takes 400K tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy3
GPT-5.4 Nano $0.0008, Hy3 $0.0004 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5.4 Nano has the number, $0.14.
Directional only
At a glance
- Coding index
- GPT-5.4 Nano 44.8 from 2 benchmarks, Hy3 not benchmarked.
- Widest category
- Math, 11.7 points apart on average over 2 shared benchmarks.
- Widest single gap
- LMArena Text: Chinese: GPT-5.4 Nano 1391 against Hy3 1485, 15.7 points apart on a 0 to 100 scale.
- Output price
- GPT-5.4 Nano $1.25 against Hy3 $0.528 per million output tokens, 2.4x apart.
- Context window
- GPT-5.4 Nano takes 400K tokens, Hy3 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-5.4 Nano | Bars | Hy3 | Ties |
|---|---|---|---|---|
| Coding | 1 | 3 | ||
| Reasoning | 0 | 5 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 4 | ||
| General | 0 | 18 | ||
| All shared boards | 1 | 32 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index13
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| LMArena WebDev | Not measured | 1507 | Only Hy3 measured |
| AA Coding Index | 56.1 | Not measured | Only GPT-5.4 Nano measured |
| LiveCodeBench (Vals AI) | 84.0% | Not measured | Only GPT-5.4 Nano measured |
| SWE-bench Verified (Vals AI) | 69.8% | Not measured | Only GPT-5.4 Nano measured |
| Vibe Code Bench (Vals AI) | 26.1% | Not measured | Only GPT-5.4 Nano measured |
| Code migration (Vals AI) | 14.5% | Not measured | Only GPT-5.4 Nano measured |
| Terminal-Bench 2.1 (Vals AI) | 41.6% | Not measured | Only GPT-5.4 Nano measured |
| SciCode | 47.2% | 48.6% | Hy3 |
| Terminal-Bench Hard (Artificial Analysis run) | 42.4% | Not measured | Only GPT-5.4 Nano measured |
| WeirdML | 49.2% | Not measured | Only GPT-5.4 Nano measured |
| ALE-Bench | 1004.5 | 838.6 | GPT-5.4 Nano |
| LMArena Text: coding | 1404 | 1466 | Hy3 |
| LMArena Text: software and IT | 1402 | 1463 | Hy3 |
| LMArena WebDev: React | Not measured | 1511 | Only Hy3 measured |
| LMArena WebDev: HTML | Not measured | 1487 | Only Hy3 measured |
| Neither model has a result on the other 24 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.7 | Not measured | Only GPT-5.4 Nano measured |
| Epoch Capabilities Index | 146.0 | Not measured | Only GPT-5.4 Nano measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| AA Agentic Index | 16.0 | Not measured | Only GPT-5.4 Nano measured |
| Finance Agent (Vals AI) | 38.2% | Not measured | Only GPT-5.4 Nano measured |
| Tax Agent Bench (Vals AI) | 5.0% | Not measured | Only GPT-5.4 Nano measured |
| Legal Research Bench (Vals AI) | 6.3% | Not measured | Only GPT-5.4 Nano measured |
| Excel Modeling Benchmark (Vals AI) | 44.8% | Not measured | Only GPT-5.4 Nano measured |
| τ²-Bench Telecom | 76.0% | Not measured | Only GPT-5.4 Nano measured |
| Neither model has a result on the other 35 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| GPQA Diamond | 77.5% | Not measured | Only GPT-5.4 Nano measured |
| MMMU Pro (Vals AI) | 73.6% | Not measured | Only GPT-5.4 Nano measured |
| GPQA Diamond (Artificial Analysis run) | 81.7% | 89.7% | Hy3 |
| Humanity's Last Exam (Artificial Analysis run) | 28.3% | 33.5% | Hy3 |
| GPQA Diamond (Epoch AI run) | 78.5% | Not measured | Only GPT-5.4 Nano measured |
| ARC-AGI | 51.5% | Not measured | Only GPT-5.4 Nano measured |
| ARC-AGI-2 | 5.7% | Not measured | Only GPT-5.4 Nano measured |
| Chess puzzles | 45.0% | Not measured | Only GPT-5.4 Nano measured |
| Mystery game puzzles | 9.0% | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: hard prompts | 1381 | 1447 | Hy3 |
| LMArena Text: expert | 1396 | 1461 | Hy3 |
| LMArena Text: hard prompts in English | 1387 | 1455 | Hy3 |
| LMArena Vision: diagrams | 1210 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: homework | 1225 | Not measured | Only GPT-5.4 Nano measured |
| Kagi LLM Benchmark | 39.7% | Not measured | Only GPT-5.4 Nano measured |
| Neither model has a result on the other 7 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| AIME | 88.8% | Not measured | Only GPT-5.4 Nano measured |
| OTIS Mock AIME | 87.8% | Not measured | Only GPT-5.4 Nano measured |
| FrontierMath, tiers 1 to 3 | 44.9% | Not measured | Only GPT-5.4 Nano measured |
| FrontierMath, tier 4 | 12.2% | Not measured | Only GPT-5.4 Nano measured |
| ProofBench | 5.0% | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: math | 1405 | 1473 | Hy3 |
| LMArena Text: mathematical work | 1409 | 1482 | Hy3 |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| MMLU Pro | 77.2% | Not measured | Only GPT-5.4 Nano measured |
| LegalBench (Vals AI) | 77.9% | Not measured | Only GPT-5.4 Nano measured |
| CaseLaw (Vals AI) | 51.9% | Not measured | Only GPT-5.4 Nano measured |
| CorpFin (Vals AI) | 61.2% | Not measured | Only GPT-5.4 Nano measured |
| TaxEval (Vals AI) | 67.4% | Not measured | Only GPT-5.4 Nano measured |
| MortgageTax (Vals AI) | 59.1% | Not measured | Only GPT-5.4 Nano measured |
| MedCode (Vals AI) | 41.0% | Not measured | Only GPT-5.4 Nano measured |
| SimpleQA Verified | 11.7% | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: science | 1374 | 1457 | Hy3 |
| LMArena Text: medicine and healthcare | 1369 | 1452 | Hy3 |
| LMArena Text: legal and government | 1368 | 1422 | Hy3 |
| LMArena Text: business and finance | 1366 | 1424 | Hy3 |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1230 | Only Hy3 measured |
| Design Arena: UI components | Not measured | 1204 | Only Hy3 measured |
| Design Arena: game dev | Not measured | 1208 | Only Hy3 measured |
| Design Arena: data viz | Not measured | 1165 | Only Hy3 measured |
| Design Arena: 3D | Not measured | 1232 | Only Hy3 measured |
| Design Arena: all code categories | Not measured | 1226 | Only Hy3 measured |
| Design Arena: SVG | Not measured | 1200 | Only Hy3 measured |
| 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 index018
| Benchmark | GPT-5.4 Nano | Hy3 | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 38.1% | Not measured | Only GPT-5.4 Nano measured |
| MedScribe (Vals AI) | 77.1% | Not measured | Only GPT-5.4 Nano measured |
| Vals Multimodal Index | 47.6% | Not measured | Only GPT-5.4 Nano measured |
| IFBench | 75.9% | Not measured | Only GPT-5.4 Nano measured |
| Long context reasoning (AA-LCR) | 76.7% | 79.0% | Hy3 |
| DTBench | 80.3% | Not measured | Only GPT-5.4 Nano measured |
| LMCA | 36.9% | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text | 1372 | 1438 | Hy3 |
| LMArena Text: instruction following | 1362 | 1426 | Hy3 |
| LMArena Text: creative writing | 1315 | 1403 | Hy3 |
| LMArena Vision | 1196 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: English | 1381 | 1448 | Hy3 |
| LMArena Text: non-English | 1359 | 1426 | Hy3 |
| LMArena Text: Chinese | 1391 | 1485 | Hy3 |
| LMArena Text: German | 1368 | 1449 | Hy3 |
| LMArena Text: Japanese | 1343 | 1397 | Hy3 |
| LMArena Text: Korean | 1321 | 1396 | Hy3 |
| LMArena Text: French | 1395 | 1456 | Hy3 |
| LMArena Text: Spanish | 1371 | 1457 | Hy3 |
| LMArena Text: Russian | 1363 | 1432 | Hy3 |
| LMArena Text: Polish | 1366 | 1447 | Hy3 |
| LMArena Text: multi-turn | 1382 | 1436 | Hy3 |
| LMArena Text: longer queries | 1365 | 1442 | Hy3 |
| LMArena Text: writing and language | 1338 | 1414 | Hy3 |
| LMArena Text: entertainment and media | 1325 | 1400 | Hy3 |
| LMArena Vision: English | 1202 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: Chinese | 1232 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: reading text | 1204 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: humor | 1161 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: recognizing things | 1141 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Vision: creative writing | 1159 | Not measured | Only GPT-5.4 Nano measured |
| Neither model has a result on the other 15 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.4 Nano | 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 | OpenAI GPT-5.4 Nano | Tencent Hy3 |
|---|---|---|
| Maker | OpenAI | Tencent |
| Weights | No public weights listed | Open weights |
| Released | March 17, 2026 | July 6, 2026 |
| Context window | 400K | 262K |
| Input $/Mtok | $0.20 | $0.132 |
| Output $/Mtok | $1.25 | $0.528 |
| Coding index | 44.8 | Not benchmarked |
| Independent measurements behind it | 2 | 1 |
| Cost per solved task | $0.14 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5.4 Nano better than Hy3 for coding?
Only GPT-5.4 Nano has a coding score, 44.8 on the Vibe Coding Index; Hy3 has none, so the two cannot be ranked on coding. Pick Hy3 for output tokens 2.4x cheaper ($0.528 vs $1.25 per million) and work measured by SciCode or LMArena Text: coding, where it leads.
Which is cheaper, GPT-5.4 Nano or Hy3?
GPT-5.4 Nano costs $0.20 per million input tokens and $1.25 per million output tokens; Hy3 costs $0.132 and $0.528. Cost per solved task is known for GPT-5.4 Nano only.
Which has the bigger context window, GPT-5.4 Nano or Hy3?
GPT-5.4 Nano, with 400K tokens against 262K.
How many benchmarks is the GPT-5.4 Nano vs Hy3 comparison based on?
33 benchmarks have a result for both models as of October 22, 2026. GPT-5.4 Nano's coding index rests on 2 independent benchmarks and Hy3's on 1. Where only one model has a result, the page says so instead of naming a winner.