Astra vs Claude Opus 5: benchmark scores and double the token price
The post lists Astra's ARC-AGI-3 and FrontierMath results against Opus 5's benchmarks, notes Astra's 2x token pricing, and argues for a same-screen, same-goal comparison.
GPT-6 Astra vs Claude Opus 5 is getting ridiculous Astra : → 99.9% on ARC-AGI-3 → 98% on FrontierMath Tier 4 → $10 input / $50 output per 1M tokens Claude Opus 5 : → SOTA on Frontier-Bench → SOTA on GDPval-AA → ~1.5x the next-best model on AutomationBench at the same cost → $5 input / $25 output per 1M tokens so Astra costs exactly 2x more. and now both are being pushed far beyond chat: coding, agents, computer use, long-running workflows and real-time environments. the numbers are already absurd. but the more interesting comparison is becoming much simpler: give both models the exact same screen. the exact same controls. the exact same objective. and see which one actually performs better. GPT-6 Astra vs Claude Opus 5. $50 vs $25 per million output tokens. this might be the benchmark that actually matters.
This post quotes another post: “GPT-6 Astra announced with a claim to do anything on a computer”