# GPT-5.2 vs GPT-6 Astra

> GPT-6 Astra is the stronger model overall, scoring 70.8 to 54.1 on the Noometry Index. GPT-5.2 costs 4.2× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/gpt-5-2-vs-gpt-6-astra
- Last updated: 2026-10-12
- Shared benchmarks: 44

## Summary

- They share 44 benchmarks with published results for both. GPT-5.2 scores higher in 0 categories and GPT-6 Astra in 10 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-6 Astra leads 85.1 to 50.2.
- The biggest single-benchmark swing is ProofBench: 15% for GPT-5.2 and 99% for GPT-6 Astra.
- GPT-5.2 is cheaper at $1.75 / $14 per million input/output tokens, against $10 / $50 for GPT-6 Astra.
- GPT-6 Astra accepts more context: 1.05M tokens versus 400K.

## Snapshot

| | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| Provider | OpenAI | OpenAI |
| Noometry Index | 54.1 | 70.8 |
| Rank | 34 | 1 |
| Context | 400K | 1.05M |
| Input $/M | $1.75 | $10 |
| Output $/M | $14 | $50 |
| Weights | Proprietary | Proprietary |

## Coding

- GPT-5.2: 51.6 (#37)
- GPT-6 Astra: 73.7 (#2)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena WebDev | 1416 | 1786 |
| GSO | 27.4% | 79.4% |
| WeirdML | 72.2% | 93.6% |
| LMArena Coding | 1447 | 1487 |
| ALE-Bench | 1,294 | 2,951 |
| SWE-bench Verified | 73.8% | — |
| DeepSWE | — | 74.1% |
| FrontierCode | — | 53.3% |
| SWE-bench Verified (bash only) | 72.8% | — |
| SWE-bench Multilingual | 66.7% | — |
| FrontierSWE | — | 65.5% |
| SciCode | — | 56.5% |
| MirrorCode | — | 46.7% |
| AlgoTune | 2.05 | — |

## Agentic & Tool Use

- GPT-5.2: 40.2 (#24)
- GPT-6 Astra: 52.9 (#3)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| Remote Labor Index | 2.5% | 20.8% |
| Vending-Bench 2 | 3,591 | 15,515 |
| Terminal-Bench | 64.9% | — |
| APEX-Agents | — | 64.7% |
| Berkeley Function Calling Leaderboard | 55.9% | — |
| GDPval | 49.7% | — |
| τ²-bench Airline | 83% | — |
| τ²-bench Banking | 32.2% | — |
| τ²-bench Retail | 81.6% | — |
| τ²-bench Telecom | 89.7% | — |
| DeepResearch Bench | 41.1% | — |
| BALROG | — | 68.3% |
| GDP.pdf | — | 34.2% |
| LMArena Search | 1207 | — |
| METR Time Horizons | 75.3% | — |

## Reasoning

- GPT-5.2: 50.2 (#35)
- GPT-6 Astra: 85.1 (#1)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| ARC-AGI-2 | 52.9% | 95% |
| NYT Connections (extended) | 83.6% | 98.1% |
| ARC-AGI-1 | 86.2% | 98.5% |
| Chess Puzzles | 49% | 72% |
| EBR-Bench | 23% | 76.2% |
| LMArena Hard Prompts | 1445 | 1462 |
| Mystery Game Puzzles | 23% | 84% |
| DTBench | 90.9% | 97.3% |
| LMCA | 43.9% | 64.4% |
| Epoch Capabilities Index | 153.45 | 166.45 |
| SimpleBench | 45.8% | — |
| Kagi LLM Benchmark | 73.3% | — |
| CritPt | — | 31.7% |
| EnigmaEval | 10.4% | — |
| Bench to the Future 3 | — | 0.14 |
| ForecastBench | 60.1 | — |

## Math

- GPT-5.2: 60.0 (#38)
- GPT-6 Astra: 93.5 (#2)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| FrontierMath (Tiers 1-3) | 67.4% | 93.7% |
| FrontierMath Tier 4 | 31.7% | 97.6% |
| OTIS Mock AIME 2024-2025 | 96.1% | 100% |
| ProofBench | 15% | 99% |
| LMArena Math | 1440 | 1465 |
| MathArena Final-Answer Competitions | 72% | — |
| FrontierMath (Feb 2025 set) | 40.7% | — |
| FrontierMath Erdős | — | 2.9% |
| FrontierMath Tier 4 (v1) | 18.8% | — |

## Knowledge

- GPT-5.2: 59.3 (#32)
- GPT-6 Astra: 75.3 (#1)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| GPQA Diamond | 91.4% | 95.8% |
| Humanity's Last Exam | 27.8% | 54.8% |
| SimpleQA Verified | 37.1% | 75.6% |
| Vectara Hallucination Rate | 8.4% | 8.7% |
| LMArena Expert | 1445 | 1483 |

## Multimodal

- GPT-5.2: 51.3 (#7)
- GPT-6 Astra: 55.0 (#3)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena Vision | 1268 | 1281 |
| Furniture Assembly | 38.3% | 80% |
| LMArena Document | 1405 | 1468 |
| VPCT | 84% | — |
| Blueprint-Bench 2 | — | 49.7% |

## Multilingual

- GPT-5.2: 53.4 (#67)
- GPT-6 Astra: 53.7 (#61)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena Non-English | 1425 | 1430 |
| LMArena Chinese | 1460 | 1484 |
| LMArena French | 1455 | 1456 |
| LMArena German | 1448 | 1440 |
| LMArena Japanese | 1420 | 1379 |
| LMArena Korean | 1392 | 1426 |
| LMArena Russian | 1440 | 1436 |
| LMArena Spanish | 1433 | 1407 |

## Instruction Following

- GPT-5.2: 74.7 (#89)
- GPT-6 Astra: 76.3 (#44)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena Instruction Following | 1417 | 1450 |

## Long Context

- GPT-5.2: 44.0 (#78)
- GPT-6 Astra: 44.5 (#62)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena Longer Query | 1428 | 1456 |
| CL-bench | 18.2% | — |

## Writing & Preference

- GPT-5.2: 66.8 (#32)
- GPT-6 Astra: 75.3 (#7)

| Benchmark | GPT-5.2 | GPT-6 Astra |
|---|---|---|
| LMArena Text | 1439 | 1441 |
| LMArena Creative Writing | 1401 | 1418 |
| EQ-Bench Creative Writing | 1703 | 2173 |
| LMArena Multi-Turn | 1458 | 1448 |

## FAQ

### Is GPT-5.2 better than GPT-6 Astra?

GPT-6 Astra is the stronger model overall, scoring 70.8 to 54.1 on the Noometry Index. GPT-5.2 costs 4.2× less per token, which makes it the better buy when GPT-6 Astra's lead doesn't matter for your workload.

### Which is cheaper, GPT-5.2 or GPT-6 Astra?

GPT-5.2 is cheaper. It lists at $1.75 per million input tokens and $14 per million output tokens; GPT-6 Astra lists at $10 and $50.

### Is GPT-5.2 or GPT-6 Astra better for coding?

GPT-6 Astra scores higher on coding benchmarks: 73.7 versus 51.6 in the Noometry coding category.

### Which has the bigger context window?

GPT-6 Astra does, with 1.05M tokens against 400K.

### How many benchmarks do GPT-5.2 and GPT-6 Astra share?

44 benchmarks have published results for both models. GPT-5.2 has 67 scored results on Noometry and GPT-6 Astra has 56.