GPT-5.4 Nano vs Hy3 preview
Hy3 preview leads GPT-5.4 Nano on the Artificial Analysis Intelligence Index, 25.3 to 20.7, and Hy3 preview costs 2.1x less per output token ($0.60 vs $1.25 per million). Updated October 22, 2026; 33 benchmarks have a result for both.
GPT-5.4 Nano or Hy3 preview: which to pick
Only GPT-5.4 Nano has a coding score, 44.8 on the Vibe Coding Index; Hy3 preview has none, so the two cannot be ranked on coding.
Pick Hy3 preview for output tokens 2.1x cheaper ($0.60 vs $1.25 per million) and work measured by AA Coding Index or LMArena Text: coding, where it leads.
- Shared results
- 33
- GPT-5.4 Nano only
- 52
- Hy3 preview only
- 3
- Like-for-like categories
- 7 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
Hy3 preview
Hy3 preview leads, 25.3 against 20.7.
Same board, both measured
Agentic work
AA Agentic Index
Hy3 preview
Hy3 preview leads, 24.1 against 16.0.
Same board, both measured
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
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 preview
GPT-5.4 Nano $0.0008, Hy3 preview $0.0005 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 preview not benchmarked.
- Widest category
- Agentic work, 12.4 points apart on average over 2 shared benchmarks.
- Widest single gap
- τ²-Bench Telecom: GPT-5.4 Nano 76.0% against Hy3 preview 92.7%, 16.7 points apart on a 0 to 100 scale.
- Output price
- GPT-5.4 Nano $1.25 against Hy3 preview $0.60 per million output tokens, 2.1x apart.
- Context window
- GPT-5.4 Nano takes 400K 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-5.4 Nano | Bars | Hy3 preview | Ties |
|---|---|---|---|---|
| Coding | 1 | 3 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 2 | ||
| Reasoning | 1 | 4 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 4 | ||
| General | 2 | 13 | ||
| All shared boards | 4 | 29 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index13
| Benchmark | GPT-5.4 Nano | Hy3 preview | Leader |
|---|---|---|---|
| LMArena WebDev | Not measured | 1357 | Only Hy3 preview measured |
| AA Coding Index | 56.1 | 58.8 | Hy3 preview |
| 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% | Not measured | Only GPT-5.4 Nano measured |
| Terminal-Bench Hard (Artificial Analysis run) | 42.4% | 34.1% | GPT-5.4 Nano |
| WeirdML | 49.2% | Not measured | Only GPT-5.4 Nano measured |
| ALE-Bench | 1004.5 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: coding | 1404 | 1421 | Hy3 preview |
| LMArena Text: software and IT | 1402 | 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 24 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5.4 Nano | Hy3 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.7 | 25.3 | Hy3 preview |
| Epoch Capabilities Index | 146.0 | Not measured | Only GPT-5.4 Nano measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5.4 Nano | Hy3 preview | Leader |
|---|---|---|---|
| AA Agentic Index | 16.0 | 24.1 | Hy3 preview |
| 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% | 92.7% | Hy3 preview |
| Neither model has a result on the other 35 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | GPT-5.4 Nano | Hy3 preview | 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% | 86.7% | Hy3 preview |
| Humanity's Last Exam (Artificial Analysis run) | 28.3% | 27.8% | GPT-5.4 Nano |
| 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 | 1413 | Hy3 preview |
| LMArena Text: expert | 1396 | 1422 | Hy3 preview |
| LMArena Text: hard prompts in English | 1387 | 1416 | Hy3 preview |
| 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 preview | 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 | 1424 | Hy3 preview |
| LMArena Text: mathematical work | 1409 | 1442 | Hy3 preview |
| 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 preview | 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 | 1416 | Hy3 preview |
| LMArena Text: medicine and healthcare | 1369 | 1405 | Hy3 preview |
| LMArena Text: legal and government | 1368 | 1430 | Hy3 preview |
| LMArena Text: business and finance | 1366 | 1397 | Hy3 preview |
| Neither model has a result on the other 6 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5.4 Nano | Hy3 preview | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index213
| Benchmark | GPT-5.4 Nano | Hy3 preview | 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% | 63.1% | GPT-5.4 Nano |
| Long context reasoning (AA-LCR) | 76.7% | 64.7% | GPT-5.4 Nano |
| 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 | 1401 | Hy3 preview |
| LMArena Text: instruction following | 1362 | 1383 | Hy3 preview |
| LMArena Text: creative writing | 1315 | 1339 | Hy3 preview |
| LMArena Vision | 1196 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: English | 1381 | 1409 | Hy3 preview |
| LMArena Text: non-English | 1359 | 1390 | Hy3 preview |
| LMArena Text: Chinese | 1391 | 1435 | Hy3 preview |
| LMArena Text: German | 1368 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: Japanese | 1343 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: Korean | 1321 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: French | 1395 | 1425 | Hy3 preview |
| LMArena Text: Spanish | 1371 | 1399 | Hy3 preview |
| LMArena Text: Russian | 1363 | 1399 | Hy3 preview |
| LMArena Text: Polish | 1366 | Not measured | Only GPT-5.4 Nano measured |
| LMArena Text: multi-turn | 1382 | 1411 | Hy3 preview |
| LMArena Text: longer queries | 1365 | 1402 | Hy3 preview |
| LMArena Text: writing and language | 1338 | 1370 | Hy3 preview |
| LMArena Text: entertainment and media | 1325 | 1352 | Hy3 preview |
| 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 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
Hy3 preview costs less
Repository review
50K input + 3K output tokens
Hy3 preview costs less
Agent loop
200K input + 10K output tokens
Hy3 preview 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 preview |
|---|---|---|
| Maker | OpenAI | Tencent |
| Weights | No public weights listed | Open weights |
| Released | March 17, 2026 | April 22, 2026 |
| Context window | 400K | 262K |
| Input $/Mtok | $0.20 | $0.18 |
| Output $/Mtok | $1.25 | $0.60 |
| 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 preview for coding?
Only GPT-5.4 Nano has a coding score, 44.8 on the Vibe Coding Index; Hy3 preview has none, so the two cannot be ranked on coding. Pick Hy3 preview for output tokens 2.1x cheaper ($0.60 vs $1.25 per million) and work measured by AA Coding Index or LMArena Text: coding, where it leads.
Which is cheaper, GPT-5.4 Nano or Hy3 preview?
GPT-5.4 Nano costs $0.20 per million input tokens and $1.25 per million output tokens; Hy3 preview costs $0.18 and $0.60. Cost per solved task is known for GPT-5.4 Nano only.
Which has the bigger context window, GPT-5.4 Nano or Hy3 preview?
GPT-5.4 Nano, with 400K tokens against 262K.
How many benchmarks is the GPT-5.4 Nano vs Hy3 preview 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 preview's on 1. Where only one model has a result, the page says so instead of naming a winner.