GPT-6 Astra vs. GPT-6.2 Astra
The new version GPT-6.2 Astra against the older GPT-6 Astra on the same 5 zero-shot prompts – what changed in function, looks, cost and time.
Short answer
- In our test GPT-6 Astra had the better result in 1 and GPT-6.2 Astra in 3 of 5 prompts, 1 tied.
- Ran on the first attempt: GPT-6 Astra 5/5, GPT-6.2 Astra 4/5.
- Average score: GPT-6 Astra 93/100, GPT-6.2 Astra 77/100 (a broken output = 0).
- Why GPT-6.2 Astra won more prompts but has a lower average: its original output (GTA-style game) did not run and counts as zero.
- Cost of all prompts: GPT-6 Astra $7.65, GPT-6.2 Astra $2.38 – GPT-6 Astra is 3.2× more expensive.
- Total generation time: GPT-6 Astra 39 min 41 s, GPT-6.2 Astra 12 min 42 s.
- GPT-6.2 Astra is better at: Solar System, Monitor test, New York & Central Park map; worse at: GTA-style game.
Tested 15 Oct 2026 – 18 Nov 2026, one attempt per prompt and model; the table shows the original outputs (a fixed copy is listed separately). This is the result of our test, not a general measure of the models' reliability. Time = from sending the prompt to the last token of the answer, thinking included (for runs in Claude Code the whole task incl. writing the file), fixes excluded. Cost = tokens × the public API price list as of the date at the model, fixes excluded; ≈ = an estimate when only the token total is known. It is not the amount actually paid – subscriptions, caching or discounts change it.
Editor's note
GPT-6.2 Astra met every requirement (8/8) on four prompts in this test, but its GTA game could not be played on the first attempt (a bug in the engine sound), so it scores 0 there; the fixed copy by the same model met 8/8. GPT-6 Astra had all five outputs working on the first attempt. The new version is much faster and cheaper: about 13 versus 40 minutes and $2.38 versus $7.65 in total.
Results per prompt
| Prompt | GPT-6 Astra | GPT-6.2 Astra | Better result | Cost A / B | Time A / B |
|---|---|---|---|---|---|
| Solar System | 8/8 Score 90/100 | 8/8 Score 96/100 | GPT-6.2 Astra | $1.04 $0.54 | 4 min 25 s 2 min 02 s |
| GTA-style game | 8/8 Score 92/100 | broken original = 0 points fixed copy: 8/8 | GPT-6 Astra | $2.06 $0.65 | 11 min 09 s 3 min 27 s |
| Monitor test | 8/8 Score 95/100 | 8/8 Score 96/100 | GPT-6.2 Astra | $0.78 $0.40 | 4 min 18 s 1 min 55 s |
| New York & Central Park map | 8/8 Score 91/100 | 8/8 Score 97/100 | GPT-6.2 Astra | $1.50 $0.43 | 9 min 17 s 3 min 01 s |
| Chess with an engine | 8/8 Score 97/100 | 8/8 Score 97/100 | tie | $2.26 $0.36 | 10 min 32 s 2 min 17 s |
Better result = higher overall score of the original output (60 % requirements, 30 % UX / UI, 10 % technical health). Cost by the API price list as of the date at the model, ≈ = estimate. Time = from sending the prompt to the last token of the answer, thinking included (for runs in Claude Code the whole task incl. writing the file), fixes excluded. Cost = tokens × the public API price list as of the date at the model, fixes excluded; ≈ = an estimate when only the token total is known. It is not the amount actually paid – subscriptions, caching or discounts change it.
Previews and reviewers' verdicts
Output previews – click to open them live. The text under a preview is the summary of the published review.
Solar System
Score 90/100
Compact but very complete raw-WebGL orrery with real orbital elements and nice procedural planets. All requirements are implemented and wired up.
Score 96/100
An elegant orrery with its own 3D projection and shading, accurate orbits for today's date, time controls and a true-scale switch.
GTA-style game
Score 92/100
Polished, complete top-down GTA clone where every checklist item is implemented. The city is a simple regular grid but meets the requirements.
fixed copy: 8/8
The original output did not run and counts as 0 points. The preview and verdict below belong to a fixed copy by the same model (8/8) – it does not count in the comparison.
Verdict of the fixed copy: Fixed version: a compact, stylish game with every requested system – 3 cars, damage, police, 2 mission types and a minimap.
Monitor test
Score 95/100
Polished, complete monitor test suite with 12 tests, device-pixel patterns and good UX. All checklist items are implemented.
Score 96/100
A complete, visually polished suite of 12 tests with exact physical-pixel rendering and a clear explanation for every test.
New York & Central Park map
Score 91/100
High-quality illustrated Manhattan map with detailed Central Park, 55 landmarks and full navigation. All checklist items are implemented.
Score 97/100
A compact, stylish atlas-like illustrated map: rotated grid, detailed Central Park, 45 places with symbols, search and a guided tour.
Chess with an engine
Score 97/100
Compact but complete implementation with correct rules, worker-based alpha-beta engine and perft self-test; engine is slowish due to copy-make.
Score 97/100
A compact but complete chess app: exact rules (perft 6/6) including FIDE draw claims, an engine in a Web Worker with three levels and an elegant board.
The prompt and context
- GPT-6 Astra model profile (OpenAI)
- GPT-6.2 Astra model profile (OpenAI)
- Model leaderboard · Methodology
More two-model comparisons
- GPT-6 Astra vs. Claude Haiku 5.5
- Claude Opus 5.5 vs. Claude Fable 5.2
- Claude Opus 5.5 vs. GPT-6 Astra
- DeepSeek V4.1 Flash 0910 vs. DeepSeek V4.1 Flash 1017
- Kimi K3 vs. Kimi K3.1 Pro
- Grok 4.6 vs. Grok 4.7
- GPT-5.6 Terra vs. GPT-6.1 Terra
- Mistral Large 4 Preview vs. Mistral Large 4 1029
- GPT-6.1 Sol vs. GPT-6.2 Sol