GPT-6 Astra experimental context management breaks 270k token cap but at extra cost

On September 8, Brandon Galang shared an experimental trick for GPT-6 Astra in Codex: enabling the experimental context management setting hidden in an inconspicuous corner of the client UI and raising the max token limit lets the model manage context on its own by "leaving notes to itself," making contexts beyond the previous 270k token limit more practical. The tip gained attention after Paulo Portella verified and reshared it.

Confirmed

• With experimental context management enabled and the limit raised, GPT-6 Astra can manage context by "leaving notes to itself," breaking past the previous 270k limit
• Brandon Galang corrected himself: context beyond roughly 272k tokens is not billed at the standard rate and incurs higher charges; he advises against doing this
• The setting works only in the Codex client and does not apply to the API

Why it matters

This is an early real-world test of GPT-6 Astra's long-context capabilities: experimental context management offers a new way to handle extremely large contexts, but the added billing threshold means the "cheap" context practically available to ordinary users still caps at around 270k—going beyond that is more of a costly experiment.

2026-09-08 ~ 2026-09-08 · 5 related posts

• [source] Pro tip: enable experimental context management for gpt-6-astra in Codex, 400k tokens sweet spot — brandon_galang · 2026-09-08
• Codex's hidden experimental context management setting charges more above ~270k tokens — pauloportella_ · 2026-09-08
• [source] Brandon concedes: GPT-6 Astra does charge more above ~270k context — brandon_galang · 2026-09-08
• GPT-6 Astra's experimental context management allows higher token limits, but exceeding 272k incurs extra charges — brandon_galang · 2026-09-08
• [source] The context management setting is hidden and Codex-only, not available via API — brandon_galang · 2026-09-08