# OpenAI GPT-6.2 Astra

As of November 22, 2026, OpenAI GPT-6.2 Astra ranks #5 of 120 models overall with 86.4 ±5.4 on the LLMBenchmark Index. GPT-6.2 Astra ranks #1 of 125 in reasoning, its strongest category. OpenAI GPT-6.2 Astra costs $8.00 per million input tokens and $40.00 per million output tokens. GPT-6.2 Astra was released on November 17, 2026.

Page: https://llmbenchmark.ai/models/gpt-6-2-astra
Data from November 22, 2026. Every figure below names the board it was read from and the day.

- Maker: OpenAI
- Released: November 17, 2026
- Price per million input tokens: $8.00
- Price per million output tokens: $40.00
- Context window: 1.05M tokens
- Overall rank: 5 of 120, by the LLMBenchmark Index
- LLMBenchmark Index: 86.4 ±5.4 (90% margin, from 7 of 7 inputs)
- Artificial Analysis Intelligence Index: 62.1
- Vibe Coding Index: 87.0 (from 4 of 6 coding benchmarks)

Benchmark results (76 of 222):

- LiveBench coding: 91.9% (source: LiveBench, read November 22, 2026)
- LMArena WebDev: 1766 Elo (Max) (source: LMArena WebDev, read November 22, 2026)
- Vals Index: 67.7% (source: Vals Index, read November 22, 2026)
- AA Intelligence Index: 62.1 (source: Artificial Analysis (via OpenRouter), read November 22, 2026)
- Design Arena: websites: 1333 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: UI components: 1341 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: game dev: 1501 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: 3D: 1430 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: all code categories: 1389 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: agents building web apps: 1334 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: agents building full-stack apps: 1351 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: agents building mobile apps: 1305 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: agents building Android apps: 1257 Elo (source: Design Arena, read November 22, 2026)
- Design Arena: spreadsheets: 1278 Elo (source: Design Arena, read November 22, 2026)
- LiveBench reasoning: 91.6% (source: LiveBench, read November 22, 2026)
- LiveBench data analysis: 84.1% (source: LiveBench, read November 22, 2026)
- LiveBench research: 91.5% (source: LiveBench, read November 22, 2026)
- LiveBench browser: 93.0% (source: LiveBench, read November 22, 2026)
- LiveBench long agentic: 70.8% (source: LiveBench, read November 22, 2026)
- LiveBench design: 82.8% (source: LiveBench, read November 22, 2026)
- Terminal-Bench 4 (Vals AI): 63.1% (Mini-SWE-agent) (source: Vals AI boards, read November 22, 2026)
- Vibe Code Bench (Vals AI): 89.9% (OpenHands) (source: Vals AI boards, read November 22, 2026)
- Tax Agent Bench (Vals AI): 21.8% (source: Vals AI boards, read November 22, 2026)
- Legal Research Bench (Vals AI): 42.8% (source: Vals AI boards, read November 22, 2026)
- Harvey's Legal Agent Benchmark (Vals AI): 16.3% (source: Vals AI boards, read November 22, 2026)
- BioMysteryBench (Vals AI): 80.7% (source: Vals AI boards, read November 22, 2026)
- Terminal-Bench-Science (Vals AI): 71.4% (Mini-SWE-agent) (source: Vals AI boards, read November 22, 2026)
- Vals Science Index: 68.4% (source: Vals AI boards, read November 22, 2026)
- Robo-bench (Vals AI): 32.7% (source: Vals AI boards, read November 22, 2026)
- MysteryMechanism (Vals AI): 53.1% (source: Vals AI boards, read November 22, 2026)
- Terminal-Bench 4.0 (Artificial Analysis run): 64.1% (Max) (source: Artificial Analysis, read November 22, 2026)
- FrontierCode 2.0: 49.0% (xHigh) (source: Artificial Analysis, read November 22, 2026)
- HLE-Diamond (Artificial Analysis run): 72.2% (Max) (source: Artificial Analysis, read November 22, 2026)
- AA-Omniscience: 50.0 (Max) (source: Artificial Analysis, read November 22, 2026)
- CritPt v2, hard subset (Artificial Analysis run): 71.5% (Max) (source: Artificial Analysis, read November 22, 2026)
- MirrorCode: 67.8% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- APEX-Agents: 72.2% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- GPQA Diamond (Epoch AI run): 96.8% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- ARC-AGI: 98.5% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- ARC-AGI-2: 95.8% (xHigh) (source: Epoch AI benchmarking hub, read November 22, 2026)
- SimpleBench: 77.3% (source: Epoch AI benchmarking hub, read November 22, 2026)
- Chess puzzles: 78.0% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- Mystery game puzzles: 95.0% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- EBR-bench: 92.4% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- OTIS Mock AIME: 100.0% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- FrontierMath, tiers 1 to 3: 92.6% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- FrontierMath, tier 4: 100.0% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- SimpleQA Verified: 78.5% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- Furniture assembly: 89.2% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- SpireClimb: 41.3% (Max) (source: Epoch AI benchmarking hub, read November 22, 2026)
- Epoch Capabilities Index: 172.8 (source: Epoch AI benchmarking hub, read November 22, 2026)
- LMArena Text: 1414 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: coding: 1450 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: hard prompts: 1425 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: instruction following: 1423 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: creative writing: 1417 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: English: 1417 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: non-English: 1404 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: Russian: 1437 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: multi-turn: 1425 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: longer queries: 1431 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: hard prompts in English: 1426 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: software and IT: 1443 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: science: 1415 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: legal and government: 1415 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: business and finance: 1413 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: writing and language: 1440 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Text: entertainment and media: 1413 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena WebDev: React: 1770 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena WebDev: HTML: 1854 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- LMArena Search: 1191 Elo (Max) (source: LMArena (Hugging Face dataset), read November 22, 2026)
- MathArena: 92.9% (max) (source: MathArena, read November 22, 2026)
- MultiChallenge: 73.2% (high) (source: Scale SEAL leaderboards, read November 22, 2026)
- Humanity's Last Exam (Scale run): 66.9% (high) (source: Scale SEAL leaderboards, read November 22, 2026)
- GameBench 2: 97.2 (xHigh) (source: DuelLab GameBench 2, read November 22, 2026)
- GoBench: 2493 Elo (High) (source: GoBench (Roland Gao), read November 22, 2026)
