GPT-5 Mini vs Hy3 preview
Hy3 preview leads GPT-5 Mini on the Artificial Analysis Intelligence Index, 25.3 to 20.6, and Hy3 preview costs 3.3x less per output token ($0.60 vs $2.00 per million). Updated October 22, 2026; 31 benchmarks have a result for both.
GPT-5 Mini or Hy3 preview: which to pick
Only GPT-5 Mini has a coding score, 34.2 on the Vibe Coding Index; Hy3 preview has none, so the two cannot be ranked on coding.
Pick Hy3 preview for output tokens 3.3x cheaper ($0.60 vs $2.00 per million) and work measured by Terminal-Bench Hard (Artificial Analysis run) or LMArena Text: coding, where it leads.
- Shared results
- 31
- GPT-5 Mini only
- 70
- Hy3 preview only
- 5
- 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 Mini 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 20.6.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Hy3 preview has a AA Agentic Index result, 24.1.
Directional only
Reasoning
GPQA Diamond
No call
Only GPT-5 Mini has a GPQA Diamond result, 80.3%.
Directional only
Math
AIME
No call
Only GPT-5 Mini has a AIME result, 91.5%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-5 Mini has a MMLU Pro result, 82.2%.
Directional only
Web design
Design Arena: websites
No call
Only GPT-5 Mini has a Design Arena: websites result, 1125.
Directional only
Long documents
Listed context window
GPT-5 Mini
GPT-5 Mini takes 400K tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy3 preview
GPT-5 Mini $0.0013, Hy3 preview $0.0005 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GPT-5 Mini has the number, $0.08.
Directional only
At a glance
- Coding index
- GPT-5 Mini 34.2 from 2 benchmarks, Hy3 preview not benchmarked.
- Widest category
- Agentic work, 21.6 points apart on average over 1 shared benchmark.
- Widest single gap
- τ²-Bench Telecom: GPT-5 Mini 71.1% against Hy3 preview 92.7%, 21.6 points apart on a 0 to 100 scale.
- Output price
- GPT-5 Mini $2.00 against Hy3 preview $0.60 per million output tokens, 3.3x apart.
- Context window
- GPT-5 Mini 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 Mini | Bars | Hy3 preview | Ties |
|---|---|---|---|---|
| Coding | 0 | 3 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 1 | ||
| Reasoning | 0 | 5 | ||
| Math | 0 | 2 | ||
| Knowledge | 0 | 4 | ||
| General | 2 | 13 | ||
| All shared boards | 2 | 29 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index03
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| SWE-bench Verified | 59.8% | Not measured | Only GPT-5 Mini measured |
| SWE-bench Multilingual | 39.7% | Not measured | Only GPT-5 Mini measured |
| LMArena WebDev | Not measured | 1357 | Only Hy3 preview measured |
| AA Coding Index | Not measured | 58.8 | Only Hy3 preview measured |
| LiveCodeBench (Vals AI) | 86.6% | Not measured | Only GPT-5 Mini measured |
| SWE-bench Verified (Vals AI) | 60.8% | Not measured | Only GPT-5 Mini measured |
| Vibe Code Bench (Vals AI) | 14.2% | Not measured | Only GPT-5 Mini measured |
| LiveCodeBench (Artificial Analysis run) | 83.8% | Not measured | Only GPT-5 Mini measured |
| SciCode | 39.0% | Not measured | Only GPT-5 Mini measured |
| Terminal-Bench Hard (Artificial Analysis run) | 33.3% | 34.1% | Hy3 preview |
| Terminal-Bench | 34.8% | Not measured | Only GPT-5 Mini measured |
| WeirdML | 52.7% | Not measured | Only GPT-5 Mini measured |
| LMArena Text: coding | 1406 | 1421 | Hy3 preview |
| LMArena Text: software and IT | 1394 | 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 23 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 20.6 | 25.3 | Hy3 preview |
| Epoch Capabilities Index | 145.3 | Not measured | Only GPT-5 Mini measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| AA Agentic Index | Not measured | 24.1 | Only Hy3 preview measured |
| τ²-Bench Telecom | 71.1% | 92.7% | Hy3 preview |
| Vending-Bench 2 | -$31 | Not measured | Only GPT-5 Mini measured |
| Berkeley Function Calling Leaderboard (BFCL) | 55.5% | Not measured | Only GPT-5 Mini measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index05
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| GPQA Diamond | 80.3% | Not measured | Only GPT-5 Mini measured |
| MMMU Pro (Vals AI) | 78.9% | Not measured | Only GPT-5 Mini measured |
| GPQA Diamond (Artificial Analysis run) | 82.8% | 86.7% | Hy3 preview |
| Humanity's Last Exam (Artificial Analysis run) | 21.5% | 27.8% | Hy3 preview |
| GPQA Diamond (Epoch AI run) | 75.0% | Not measured | Only GPT-5 Mini measured |
| ARC-AGI | 54.3% | Not measured | Only GPT-5 Mini measured |
| ARC-AGI-2 | 4.4% | Not measured | Only GPT-5 Mini measured |
| Chess puzzles | 24.0% | Not measured | Only GPT-5 Mini measured |
| Mystery game puzzles | 10.0% | Not measured | Only GPT-5 Mini measured |
| VPCT | 40.2% | Not measured | Only GPT-5 Mini measured |
| LMArena Text: hard prompts | 1380 | 1413 | Hy3 preview |
| LMArena Text: expert | 1379 | 1422 | Hy3 preview |
| LMArena Text: hard prompts in English | 1383 | 1416 | Hy3 preview |
| LMArena Vision: diagrams | 1202 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: homework | 1210 | Not measured | Only GPT-5 Mini measured |
| Kagi LLM Benchmark | 70.3% | Not measured | Only GPT-5 Mini measured |
| Neither model has a result on the other 6 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| AIME | 91.5% | Not measured | Only GPT-5 Mini measured |
| MATH 500 (Vals AI) | 94.8% | Not measured | Only GPT-5 Mini measured |
| MGSM (Vals AI) | 92.6% | Not measured | Only GPT-5 Mini measured |
| AA Math Index | 90.7 | Not measured | Only GPT-5 Mini measured |
| AIME 2025 (Artificial Analysis run) | 90.7% | Not measured | Only GPT-5 Mini measured |
| OTIS Mock AIME | 86.7% | Not measured | Only GPT-5 Mini measured |
| FrontierMath, tiers 1 to 3 | 46.7% | Not measured | Only GPT-5 Mini measured |
| FrontierMath, tier 4 | 12.2% | Not measured | Only GPT-5 Mini measured |
| ProofBench | 9.0% | Not measured | Only GPT-5 Mini measured |
| MATH level 5 | 97.8% | Not measured | Only GPT-5 Mini measured |
| LMArena Text: math | 1378 | 1424 | Hy3 preview |
| LMArena Text: mathematical work | 1380 | 1442 | Hy3 preview |
| Neither model has a result on the other 1 board here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| MMLU Pro | 82.2% | Not measured | Only GPT-5 Mini measured |
| LegalBench (Vals AI) | 81.8% | Not measured | Only GPT-5 Mini measured |
| CaseLaw (Vals AI) | 68.5% | Not measured | Only GPT-5 Mini measured |
| CorpFin (Vals AI) | 60.2% | Not measured | Only GPT-5 Mini measured |
| TaxEval (Vals AI) | 75.2% | Not measured | Only GPT-5 Mini measured |
| MortgageTax (Vals AI) | 66.9% | Not measured | Only GPT-5 Mini measured |
| MedQA (Vals AI) | 96.1% | Not measured | Only GPT-5 Mini measured |
| MedCode (Vals AI) | 43.0% | Not measured | Only GPT-5 Mini measured |
| MMLU Pro (Artificial Analysis run) | 83.7% | Not measured | Only GPT-5 Mini measured |
| SimpleQA Verified | 21.6% | Not measured | Only GPT-5 Mini measured |
| LMArena Text: science | 1371 | 1416 | Hy3 preview |
| LMArena Text: medicine and healthcare | 1380 | 1405 | Hy3 preview |
| LMArena Text: legal and government | 1375 | 1430 | Hy3 preview |
| LMArena Text: business and finance | 1358 | 1397 | Hy3 preview |
| FACTS Benchmark Suite | 38.6% | Not measured | Only GPT-5 Mini measured |
| Neither model has a result on the other 3 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1125 | Not measured | Only GPT-5 Mini measured |
| Design Arena: UI components | 1103 | Not measured | Only GPT-5 Mini measured |
| Design Arena: game dev | 1129 | Not measured | Only GPT-5 Mini measured |
| Design Arena: data viz | 1131 | Not measured | Only GPT-5 Mini measured |
| Design Arena: 3D | 1036 | Not measured | Only GPT-5 Mini measured |
| Design Arena: all code categories | 1114 | Not measured | Only GPT-5 Mini measured |
| Design Arena: SVG | 1091 | Not measured | Only GPT-5 Mini measured |
| Design Arena: ASCII art | 1125 | Not measured | Only GPT-5 Mini measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index213
| Benchmark | GPT-5 Mini | Hy3 preview | Leader |
|---|---|---|---|
| SAGE (Vals AI) | 43.0% | Not measured | Only GPT-5 Mini measured |
| MedScribe (Vals AI) | 80.6% | Not measured | Only GPT-5 Mini measured |
| IFBench | 75.4% | 63.1% | GPT-5 Mini |
| Long context reasoning (AA-LCR) | 72.3% | 64.7% | GPT-5 Mini |
| DTBench | 80.5% | Not measured | Only GPT-5 Mini measured |
| LMCA | 34.2% | Not measured | Only GPT-5 Mini measured |
| Fiction.LiveBench | 69.4% | Not measured | Only GPT-5 Mini measured |
| Creative writing (Lech Mazur) | 83.1% | Not measured | Only GPT-5 Mini measured |
| LMArena Text | 1373 | 1401 | Hy3 preview |
| LMArena Text: instruction following | 1357 | 1383 | Hy3 preview |
| LMArena Text: creative writing | 1325 | 1339 | Hy3 preview |
| LMArena Vision | 1202 | Not measured | Only GPT-5 Mini measured |
| LMArena Text: English | 1379 | 1409 | Hy3 preview |
| LMArena Text: non-English | 1363 | 1390 | Hy3 preview |
| LMArena Text: Chinese | 1386 | 1435 | Hy3 preview |
| LMArena Text: German | 1366 | Not measured | Only GPT-5 Mini measured |
| LMArena Text: Japanese | 1341 | Not measured | Only GPT-5 Mini measured |
| LMArena Text: Korean | 1308 | Not measured | Only GPT-5 Mini measured |
| LMArena Text: French | 1386 | 1425 | Hy3 preview |
| LMArena Text: Spanish | 1354 | 1399 | Hy3 preview |
| LMArena Text: Russian | 1362 | 1399 | Hy3 preview |
| LMArena Text: Polish | 1378 | Not measured | Only GPT-5 Mini measured |
| LMArena Text: multi-turn | 1363 | 1411 | Hy3 preview |
| LMArena Text: longer queries | 1355 | 1402 | Hy3 preview |
| LMArena Text: writing and language | 1344 | 1370 | Hy3 preview |
| LMArena Text: entertainment and media | 1340 | 1352 | Hy3 preview |
| LMArena Vision: English | 1216 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: Chinese | 1214 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: reading text | 1199 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: humor | 1205 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: captioning | 1217 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: recognizing things | 1226 | Not measured | Only GPT-5 Mini measured |
| LMArena Vision: creative writing | 1181 | Not measured | Only GPT-5 Mini measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-5 Mini | 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 Mini | Tencent Hy3 preview |
|---|---|---|
| Maker | OpenAI | Tencent |
| Weights | No public weights listed | Open weights |
| Released | August 7, 2025 | April 22, 2026 |
| Context window | 400K | 262K |
| Input $/Mtok | $0.25 | $0.18 |
| Output $/Mtok | $2.00 | $0.60 |
| Coding index | 34.2 | Not benchmarked |
| Independent measurements behind it | 2 | 1 |
| Cost per solved task | $0.08 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-5 Mini better than Hy3 preview for coding?
Only GPT-5 Mini has a coding score, 34.2 on the Vibe Coding Index; Hy3 preview has none, so the two cannot be ranked on coding. Pick Hy3 preview for output tokens 3.3x cheaper ($0.60 vs $2.00 per million) and work measured by Terminal-Bench Hard (Artificial Analysis run) or LMArena Text: coding, where it leads.
Which is cheaper, GPT-5 Mini or Hy3 preview?
GPT-5 Mini costs $0.25 per million input tokens and $2.00 per million output tokens; Hy3 preview costs $0.18 and $0.60. Cost per solved task is known for GPT-5 Mini only.
Which has the bigger context window, GPT-5 Mini or Hy3 preview?
GPT-5 Mini, with 400K tokens against 262K.
How many benchmarks is the GPT-5 Mini vs Hy3 preview comparison based on?
31 benchmarks have a result for both models as of October 22, 2026. GPT-5 Mini'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.