Hy3 preview vs Hy4
Hy3 preview costs 2.1x less per output token than Hy4 ($0.60 vs $1.2505 per million). Updated November 14, 2026; no benchmark has tested both models yet.
- Coding
- No data
- Overall
- No data
- Output $/M
- $0.60
- Coding
- 65.2
- Overall
- No data
- Output $/M
- $1.2505
Hy3 preview or Hy4: which to pick
Only Hy4 has a coding score, 65.2 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 $1.2505 per million).
No benchmark has tested both yet. Hy3 preview has results on LMArena WebDev, AA Coding Index and Terminal-Bench Hard (Artificial Analysis run); Hy4 on Terminal-Bench 4 (Vals AI), Vibe Code Bench (Vals AI) and Terminal-Bench 4.0 (Artificial Analysis run). Compare them on price and context instead.
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 Hy4 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
No call
Only Hy4 has a AA Intelligence Index result, 34.6.
Directional only
Agentic work
AA Agentic Index
No call
Only Hy3 preview has a AA Agentic Index result, 24.1.
Directional only
Web design
Design Arena: websites
No call
Only Hy4 has a Design Arena: websites result, 1251.
Directional only
General
Vals Index
No call
Only Hy4 has a Vals Index result, 59.4%.
Directional only
Long documents
Listed context window
Hy4
Hy4 takes 1.05M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy3 preview
Hy3 preview $0.0005, Hy4 $0.0010 per turn.
List prices
At a glance
- Coding index
- Hy3 preview not benchmarked, Hy4 65.2 from 2 benchmarks.
- Widest category
- None yet: no category has been tested on both models.
- Widest single gap
- None yet: no benchmark has tested both models.
- Output price
- Hy3 preview $0.60 against Hy4 $1.2505 per million output tokens, 2.1x apart.
- Context window
- Hy3 preview takes 262K tokens, Hy4 1.05M.
Benchmark by benchmark
Every board either model has a result on, a category to a fold; bars are each result on a 0 to 100 scale. Only one model is measured on each, so no row names a leader.
CodingDirectional onlySix of these boards feed the Vibe Coding Index13 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| LMArena WebDev | 1358 | Not measured | Only Hy3 preview measured |
| AA Coding Index | 58.8 | Not measured | Only Hy3 preview measured |
| Terminal-Bench 4 (Vals AI) | Not measured | 39.4% | Only Hy4 measured |
| Vibe Code Bench (Vals AI) | Not measured | 81.0% | Only Hy4 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 34.1% | Not measured | Only Hy3 preview measured |
| Terminal-Bench 4.0 (Artificial Analysis run) | Not measured | 37.4% | Only Hy4 measured |
| FrontierCode 2.0 | Not measured | 31.4% | Only Hy4 measured |
| LMArena Text: coding | 1421 | Not measured | Only Hy3 preview measured |
| LMArena Text: software and IT | 1423 | Not measured | Only Hy3 preview measured |
| LMArena WebDev: React | 1350 | Not measured | Only Hy3 preview measured |
| LMArena WebDev: HTML | 1355 | Not measured | Only Hy3 preview measured |
| GameBench 2 | Not measured | 48.7 | Only Hy4 measured |
| Terminal-Bench 4.0 | Not measured | 39.1% | Only Hy4 measured |
| Neither model has a result on the other 39 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index1 board measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| AA Intelligence Index | Not measured | 34.6 | Only Hy4 measured |
| Neither model has a result on the other 1 board here | |||
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index5 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| AA Agentic Index | 24.1 | Not measured | Only Hy3 preview measured |
| Tax Agent Bench (Vals AI) | Not measured | 34.4% | Only Hy4 measured |
| Legal Research Bench (Vals AI) | Not measured | 41.4% | Only Hy4 measured |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 12.5% | Only Hy4 measured |
| τ²-Bench Telecom | 92.7% | Not measured | Only Hy3 preview measured |
| Neither model has a result on the other 39 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index7 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 86.7% | Not measured | Only Hy3 preview measured |
| Humanity's Last Exam (Artificial Analysis run) | 27.8% | Not measured | Only Hy3 preview measured |
| HLE-Diamond (Artificial Analysis run) | Not measured | 32.7% | Only Hy4 measured |
| CritPt v2, hard subset (Artificial Analysis run) | Not measured | 39.0% | Only Hy4 measured |
| LMArena Text: hard prompts | 1413 | Not measured | Only Hy3 preview measured |
| LMArena Text: expert | 1424 | Not measured | Only Hy3 preview measured |
| LMArena Text: hard prompts in English | 1416 | Not measured | Only Hy3 preview measured |
| Neither model has a result on the other 19 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index2 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| LMArena Text: math | 1424 | Not measured | Only Hy3 preview measured |
| LMArena Text: mathematical work | 1442 | Not measured | Only Hy3 preview measured |
| Neither model has a result on the other 12 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index5 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| AA-Omniscience | Not measured | 5.4 | Only Hy4 measured |
| LMArena Text: science | 1416 | Not measured | Only Hy3 preview measured |
| LMArena Text: medicine and healthcare | 1406 | Not measured | Only Hy3 preview measured |
| LMArena Text: legal and government | 1431 | Not measured | Only Hy3 preview measured |
| LMArena Text: business and finance | 1399 | Not measured | Only Hy3 preview measured |
| Neither model has a result on the other 16 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index5 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1251 | Only Hy4 measured |
| Design Arena: UI components | Not measured | 1197 | Only Hy4 measured |
| Design Arena: game dev | Not measured | 1271 | Only Hy4 measured |
| Design Arena: 3D | Not measured | 1251 | Only Hy4 measured |
| Design Arena: all code categories | Not measured | 1254 | Only Hy4 measured |
| Neither model has a result on the other 4 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index16 boards measured
| Benchmark | Hy3 preview | Hy4 | Leader |
|---|---|---|---|
| Vals Index | Not measured | 59.4% | Only Hy4 measured |
| IFBench | 63.1% | Not measured | Only Hy3 preview measured |
| Long context reasoning (AA-LCR) | 64.7% | Not measured | Only Hy3 preview measured |
| LMArena Text | 1401 | Not measured | Only Hy3 preview measured |
| LMArena Text: instruction following | 1384 | Not measured | Only Hy3 preview measured |
| LMArena Text: creative writing | 1340 | Not measured | Only Hy3 preview measured |
| LMArena Text: English | 1409 | Not measured | Only Hy3 preview measured |
| LMArena Text: non-English | 1390 | Not measured | Only Hy3 preview measured |
| LMArena Text: Chinese | 1435 | Not measured | Only Hy3 preview measured |
| LMArena Text: French | 1422 | Not measured | Only Hy3 preview measured |
| LMArena Text: Spanish | 1401 | Not measured | Only Hy3 preview measured |
| LMArena Text: Russian | 1399 | Not measured | Only Hy3 preview measured |
| LMArena Text: multi-turn | 1410 | Not measured | Only Hy3 preview measured |
| LMArena Text: longer queries | 1402 | Not measured | Only Hy3 preview measured |
| LMArena Text: writing and language | 1371 | Not measured | Only Hy3 preview measured |
| LMArena Text: entertainment and media | 1353 | Not measured | Only Hy3 preview measured |
| Neither model has a result on the other 33 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 | Tencent Hy3 preview | Tencent Hy4 |
|---|---|---|
| Maker | Tencent | Tencent |
| Weights | Open weights | Open weights |
| Released | April 22, 2026 | November 4, 2026 |
| Context window | 262K | 1.05M |
| Input $/Mtok | $0.18 | $0.417 |
| Output $/Mtok | $0.60 | $1.2505 |
| Coding index | Not benchmarked | 65.2 |
| Independent measurements behind it | 1 | 2 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Hy3 preview better than Hy4 for coding?
Only Hy4 has a coding score, 65.2 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 $1.2505 per million).
Which is cheaper, Hy3 preview or Hy4?
Hy3 preview costs $0.18 per million input tokens and $0.60 per million output tokens; Hy4 costs $0.417 and $1.2505. Cost per solved task is known for neither of them.
Which has the bigger context window, Hy3 preview or Hy4?
Hy4, with 1.05M tokens against 262K.
Has any benchmark tested both Hy3 preview and Hy4?
No benchmark has tested both yet. Hy3 preview has results on LMArena WebDev, AA Coding Index and Terminal-Bench Hard (Artificial Analysis run); Hy4 on Terminal-Bench 4 (Vals AI), Vibe Code Bench (Vals AI) and Terminal-Bench 4.0 (Artificial Analysis run). Compare them on price and context instead. Data as of November 14, 2026.