GPT-5.4 vs GPT-6 Astra
| Spec | GPT-5.4 | GPT-6 Astra |
|---|---|---|
| Respan Index | 49.9 (#44) | 72.3 (#7) |
| Blended price per 1M | $5.63 | $20 |
| Context window | 1.1M | 1.1M |
| Input / output per 1M | $2.5 / $15 | $10 / $50 |
| Max output | 128K | 128K |
| Input | text, vision | text, vision |
| Released | 2026-03-05 | 2026-09-03 |
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | GPT-5.4 | GPT-6 Astra |
|---|---|---|
| Coding | 25.9 GPT-5.4 | 58.5 GPT-6 Astra |
| Agents | 61.9 GPT-5.4 | 78.3 GPT-6 Astra |
| Reasoning | 61.4 GPT-5.4 | 86.0 GPT-6 Astra |
| Knowledge & language | 46.9 GPT-5.4 | 75.6 GPT-6 Astra |
| Human preference | 52.9 GPT-5.4 | 60.1 GPT-6 Astra |
Benchmarks both report
GPT-6 Astra leads on 26 of the 27 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | GPT-5.4 | GPT-6 Astra |
|---|---|---|
| LMArena WebDev (LMArena) | 1462 | 1764 |
| LMArena Elo | 1475 | 1475 |
| LMArena Vision (LMArena) | 1282 | 1281 |
| MirrorCode (Epoch AI) | 23.3% | 60% |
| Chess Puzzles (Epoch AI) | 48% | 76% |
| LMArena Agent (LMArena) | 0.0267 | 0.1456 |
| APEX-Agents (Mercor) | 52.4% | 64.7% |
| DeepSWE v1.1 (Datacurve) | 51.8% | 74.1% |
| EBR-bench (Epoch AI) | 25.4% | 76.2% |
| FrontierMath Tier 4 v2 (Epoch AI) | 49% | 97.6% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 78.6% | 93.7% |
| Furniture Assembly (Epoch AI) | 37.5% | 80% |
| GPQA Diamond | 92.8% | 96% |
| LiveBench Agentic Coding (LiveBench) | 53.8% | 57.3% |
| LiveBench Coding (LiveBench) | 77.5% | 80.4% |
| LiveBench Data Analysis (LiveBench) | 79.3% | 83% |
| LiveBench Instruction Following (LiveBench) | 70.2% | 75.6% |
| LiveBench Language (LiveBench) | 82.6% | 89.4% |
| LiveBench Mathematics (LiveBench) | 94.1% | 96.8% |
| LiveBench Reasoning (LiveBench) | 88.1% | 92.7% |
| Mystery Game Puzzles (Epoch AI) | 37% | 84% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 97.8% | 100% |
| SAGE (Vals AI) | 43.3% | 46.4% |
| SimpleQA Verified | 45.1% | 75.6% |
| Vending-Bench 2 (Andon Labs) | 6144.18 | 15514.7 |
| WeirdML (Håvard Tveit Ihle) | 77.7% | 93.6% |
| GSO Opt@1 (GSO) | 30.4% | 77.5% |
FAQ
Is GPT-5.4 better than GPT-6 Astra?
GPT-6 Astra scores 72.3 on the Respan Index (#7), ahead of GPT-5.4 at 49.9 (#44). GPT-6 Astra leads on 26 of the 27 benchmarks both report.
Which is cheaper, GPT-5.4 or GPT-6 Astra?
GPT-5.4 lists at $2.5 input and $15 output per 1M tokens; GPT-6 Astra at $10 and $50. GPT-5.4 is 3.6x cheaper per token at list price.
Which has the larger context window?
Both accept 1.1M tokens of context.
Compare other models
- vsGPT-5.4 vs GPT-5.2
- vsGPT-5.5 vs GPT-5.4
- vsGPT-6.2 Astra vs GPT-6 Astra
- vsGPT-6 Astra vs Gemini 4 Argon
- vsGPT-6 Astra vs Phi-4
- vsGPT-5.4 vs Claude Opus 4.7
- vsGPT-5.4 vs DeepSeek V4 Pro
- vsGPT-5.4 vs GPT-5.4 Pro
- vsGPT-6 Astra vs Claude Fable 5.1
- vsGPT-6 Astra vs Kimi K3.1 Pro
- vsGPT-6 Astra vs Claude Sonnet 5.5