Artificial Analysis: GPT-6 Astra cuts output tokens ~10% vs GPT-5.6 Sol at max effort
peterjliu · x · 2026-09-05
Artificial Analysis reports GPT-6 Astra defines a new Pareto frontier for Intelligence Index vs output tokens per task, using 10% fewer output tokens than GPT-5.6 Sol at max effort. A commenter argues actual cost is a better measure than raw input/output token prices.
Related event: GPT-6 Astra's hidden edge: token efficiency(2 posts)→
More from Models
- Researchers find ~18k AI agents colluding to bypass sandboxes and share task answers — eliebakouch · 2026-09-05
- GPT-6 shows striking SVG generation skills in early user hands-on — ChrisGPT · 2026-09-05
- GPT Astra is a big upgrade for 3D modeling, but users burn half their quota on day one — kevinkern · 2026-09-05
- GPT 6 Astra now rolling out to ChatGPT and Codex, manual install available — oran_ge · 2026-09-05
- GPT-6 Astra base instructions run 16% longer than GPT-5.6 Sol's, cutting permission loops — WolframRvnwlf · 2026-09-05
- ChatGPT Plus Users Report Usage Limit Resets, Possibly Tied to New Model Release — Forsaken-Space1158 · 2026-09-05