GPT-6.2 Astra vs Hy4 flash
GPT-6.2 Astra leads Hy4 flash on coding, 87.0 to 44.5 on the Vibe Coding Index, and Hy4 flash costs 284.7x less per output token ($0.1405 vs $40.00 per million). Updated November 22, 2026; 12 benchmarks have a result for both.
GPT-6.2 Astra or Hy4 flash: which to pick
Pick GPT-6.2 Astra for coding: it scores 87.0 on the Vibe Coding Index against 44.5 for Hy4 flash.
Pick Hy4 flash for output tokens 284.7x cheaper ($0.1405 vs $40.00 per million) and work measured by Tax Agent Bench (Vals AI), where it leads.
- Shared results
- 12
- GPT-6.2 Astra only
- 64
- Hy4 flash only
- 0
- Like-for-like categories
- 4 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
GPT-6.2 Astra
GPT-6.2 Astra scores 87.0 against 44.5.
4 and 2 benchmarks behind the scores
Intelligence
AA Intelligence Index
No call
Only GPT-6.2 Astra has a AA Intelligence Index result, 62.1.
Directional only
Web design
Design Arena: websites
GPT-6.2 Astra
GPT-6.2 Astra leads, 1333 against 1227.
Same board, both measured
General
Vals Index
GPT-6.2 Astra
GPT-6.2 Astra leads, 67.7% against 50.8%.
Same board, both measured
Long documents
Listed context window
GPT-6.2 Astra
GPT-6.2 Astra takes 1.05M tokens against 1.05M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Hy4 flash
GPT-6.2 Astra $0.028, Hy4 flash $0.0001 per turn.
List prices
At a glance
- Coding index
- GPT-6.2 Astra 87.0 from 4 benchmarks, Hy4 flash 44.5 from 2 benchmarks.
- Widest category
- Coding, 41.9 points apart on average over 3 shared benchmarks.
- Widest single gap
- Terminal-Bench 4 (Vals AI): GPT-6.2 Astra 63.1% against Hy4 flash 15.2%, 64.0 points apart on a 0 to 100 scale.
- Output price
- GPT-6.2 Astra $40.00 against Hy4 flash $0.1405 per million output tokens, 284.7x apart.
- Context window
- GPT-6.2 Astra takes 1.05M tokens, Hy4 flash 1.05M.
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-6.2 Astra | Bars | Hy4 flash | Ties |
|---|---|---|---|---|
| Coding | 3 | 0 | ||
| Agentic work | 2 | 1 | ||
| Web design | 5 | 0 | ||
| General | 1 | 0 | ||
| All shared boards | 11 | 1 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index30
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| LiveBench coding | 91.9% | Not measured | Only GPT-6.2 Astra measured |
| LMArena WebDev | 1766 | Not measured | Only GPT-6.2 Astra measured |
| Terminal-Bench 4 (Vals AI) | 63.1% | 15.2% | GPT-6.2 Astra |
| Vibe Code Bench (Vals AI) | 89.9% | 74.8% | GPT-6.2 Astra |
| Terminal-Bench 4.0 (Artificial Analysis run) | 64.1% | Not measured | Only GPT-6.2 Astra measured |
| FrontierCode 2.0 | 49.0% | Not measured | Only GPT-6.2 Astra measured |
| MirrorCode | 67.8% | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: coding | 1450 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: software and IT | 1443 | Not measured | Only GPT-6.2 Astra measured |
| LMArena WebDev: React | 1770 | Not measured | Only GPT-6.2 Astra measured |
| LMArena WebDev: HTML | 1854 | Not measured | Only GPT-6.2 Astra measured |
| GameBench 2 | 97.2 | 50.5 | GPT-6.2 Astra |
| Neither model has a result on the other 40 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| AA Intelligence Index | 62.1 | Not measured | Only GPT-6.2 Astra measured |
| Epoch Capabilities Index | 172.8 | Not measured | Only GPT-6.2 Astra measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index21
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| Design Arena: agents building web apps | 1334 | Not measured | Only GPT-6.2 Astra measured |
| Design Arena: agents building full-stack apps | 1351 | Not measured | Only GPT-6.2 Astra measured |
| Design Arena: agents building mobile apps | 1305 | Not measured | Only GPT-6.2 Astra measured |
| Design Arena: agents building Android apps | 1257 | Not measured | Only GPT-6.2 Astra measured |
| Design Arena: spreadsheets | 1278 | Not measured | Only GPT-6.2 Astra measured |
| LiveBench research | 91.5% | Not measured | Only GPT-6.2 Astra measured |
| LiveBench browser | 93.0% | Not measured | Only GPT-6.2 Astra measured |
| LiveBench long agentic | 70.8% | Not measured | Only GPT-6.2 Astra measured |
| Tax Agent Bench (Vals AI) | 21.8% | 23.3% | Hy4 flash |
| Legal Research Bench (Vals AI) | 42.8% | 38.5% | GPT-6.2 Astra |
| Harvey's Legal Agent Benchmark (Vals AI) | 16.3% | 5.4% | GPT-6.2 Astra |
| BioMysteryBench (Vals AI) | 80.7% | Not measured | Only GPT-6.2 Astra measured |
| Terminal-Bench-Science (Vals AI) | 71.4% | Not measured | Only GPT-6.2 Astra measured |
| Vals Science Index | 68.4% | Not measured | Only GPT-6.2 Astra measured |
| Robo-bench (Vals AI) | 32.7% | Not measured | Only GPT-6.2 Astra measured |
| APEX-Agents | 72.2% | Not measured | Only GPT-6.2 Astra measured |
| SpireClimb | 41.3% | Not measured | Only GPT-6.2 Astra measured |
| LMArena Search | 1191 | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 26 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index15 boards, none shared
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| LiveBench reasoning | 91.6% | Not measured | Only GPT-6.2 Astra measured |
| MysteryMechanism (Vals AI) | 53.1% | Not measured | Only GPT-6.2 Astra measured |
| HLE-Diamond (Artificial Analysis run) | 72.2% | Not measured | Only GPT-6.2 Astra measured |
| CritPt v2, hard subset (Artificial Analysis run) | 71.5% | Not measured | Only GPT-6.2 Astra measured |
| GPQA Diamond (Epoch AI run) | 96.8% | Not measured | Only GPT-6.2 Astra measured |
| ARC-AGI | 98.5% | Not measured | Only GPT-6.2 Astra measured |
| ARC-AGI-2 | 95.8% | Not measured | Only GPT-6.2 Astra measured |
| SimpleBench | 77.3% | Not measured | Only GPT-6.2 Astra measured |
| Chess puzzles | 78.0% | Not measured | Only GPT-6.2 Astra measured |
| Mystery game puzzles | 95.0% | Not measured | Only GPT-6.2 Astra measured |
| EBR-bench | 92.4% | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: hard prompts | 1425 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: hard prompts in English | 1426 | Not measured | Only GPT-6.2 Astra measured |
| Humanity's Last Exam (Scale run) | 66.9% | Not measured | Only GPT-6.2 Astra measured |
| GoBench | 2493 | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 11 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| OTIS Mock AIME | 100.0% | Not measured | Only GPT-6.2 Astra measured |
| FrontierMath, tiers 1 to 3 | 92.6% | Not measured | Only GPT-6.2 Astra measured |
| FrontierMath, tier 4 | 100.0% | Not measured | Only GPT-6.2 Astra measured |
| MathArena | 92.9% | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 10 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| AA-Omniscience | 50.0 | Not measured | Only GPT-6.2 Astra measured |
| SimpleQA Verified | 78.5% | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: science | 1415 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: legal and government | 1415 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: business and finance | 1413 | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 16 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| Design Arena: websites | 1333 | 1227 | GPT-6.2 Astra |
| Design Arena: UI components | 1341 | 1262 | GPT-6.2 Astra |
| Design Arena: game dev | 1501 | 1202 | GPT-6.2 Astra |
| Design Arena: 3D | 1430 | 1209 | GPT-6.2 Astra |
| Design Arena: all code categories | 1389 | 1225 | GPT-6.2 Astra |
| LiveBench design | 82.8% | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 3 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | GPT-6.2 Astra | Hy4 flash | Leader |
|---|---|---|---|
| Vals Index | 67.7% | 50.8% | GPT-6.2 Astra |
| LiveBench data analysis | 84.1% | Not measured | Only GPT-6.2 Astra measured |
| Furniture assembly | 89.2% | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text | 1414 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: instruction following | 1423 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: creative writing | 1417 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: English | 1417 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: non-English | 1404 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: Russian | 1437 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: multi-turn | 1425 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: longer queries | 1431 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: writing and language | 1440 | Not measured | Only GPT-6.2 Astra measured |
| LMArena Text: entertainment and media | 1413 | Not measured | Only GPT-6.2 Astra measured |
| MultiChallenge | 73.2% | Not measured | Only GPT-6.2 Astra measured |
| Neither model has a result on the other 35 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GPT-6.2 Astra | Hy4 flash | 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
Hy4 flash costs less
Repository review
50K input + 3K output tokens
Hy4 flash costs less
Agent loop
200K input + 10K output tokens
Hy4 flash 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-6.2 Astra | Tencent Hy4 flash |
|---|---|---|
| Maker | OpenAI | Tencent |
| Weights | No public weights listed | Open weights |
| Released | November 17, 2026 | November 16, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $8.00 | $0.035 |
| Output $/Mtok | $40.00 | $0.1405 |
| Coding index | 87.0 | 44.5 |
| Independent measurements behind it | 4 | 2 |
| Cost per solved task | Not computable | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GPT-6.2 Astra better than Hy4 flash for coding?
Pick GPT-6.2 Astra for coding: it scores 87.0 on the Vibe Coding Index against 44.5 for Hy4 flash. Pick Hy4 flash for output tokens 284.7x cheaper ($0.1405 vs $40.00 per million) and work measured by Tax Agent Bench (Vals AI), where it leads.
Which is cheaper, GPT-6.2 Astra or Hy4 flash?
GPT-6.2 Astra costs $8.00 per million input tokens and $40.00 per million output tokens; Hy4 flash costs $0.035 and $0.1405. Cost per solved task is known for neither of them.
Which has the bigger context window, GPT-6.2 Astra or Hy4 flash?
GPT-6.2 Astra, with 1.05M tokens against 1.05M.
How many benchmarks is the GPT-6.2 Astra vs Hy4 flash comparison based on?
12 benchmarks have a result for both models as of November 22, 2026. GPT-6.2 Astra's coding index rests on 4 independent benchmarks and Hy4 flash's on 2. Where only one model has a result, the page says so instead of naming a winner.