Rumor: Astra Low to X-High Shows Only 1% Reasoning Difference
BLUECOW009 · x · 2026-09-06
BLUECOW009 relays an unverified claim that Astra's reasoning difference from the lowest setting (low) to the highest (x-high) is only about 1%. If true, maxing out reasoning effort would offer near-zero returns.
More from Models
- GPT-6 Astra users hit frequent 'model at capacity' errors — bytebot · 2026-09-06
- Astra User Says Weekly Token Limits Too Tight: 24-Hour Run Still Unfinished — Wooden_Drag9473 · 2026-09-06
- Researcher says OpenAI's Astra censors his newsletter, fueling safety-sandbagging concerns — sethlazar · 2026-09-06
- Deleting most Skill rules made GPT 6 better, sparking 'Skill is dead' debate — xiaohu · 2026-09-06
- Unverified: GPT-6 Astra matches GLM-5.3 max reasoning with zero reasoning tokens — BLUECOW009 · 2026-09-06
- Dev demo: OpenAI's Astra agent detects and fixes failed states via voice command — BLUECOW009 · 2026-09-06