Launch of GPT-6 Astra and price war among frontier labs
OpenAI deploys GPT-6 Astra (with Sol and Luna as variants) in the same window in which Anthropic releases Claude Opus 5.5, xAI publishes Grok 4.7 and Xiaomi launches MiMo v2.6, compressing launch cycles to minutes and shifting competition to cost per task. The capability leap —first model to reach Critical cybersecurity level under the Preparedness Framework— enables autonomous agents in production (Perplexity, Devin, Legora, Parallel) and reconfigures the marginal cost of cognitive work, with direct implications for the capital layer and for the regulation of dangerous capabilities.
Title, thesis and gravity assessment written by an AI model, without prior human review.
OpenAI deploys GPT-6 Astra with SOTA in computer use and coding in the same window as Claude Opus 5.5, Grok 4.7 and MiMo v2.6, compressing launch cycles and shifting competition to cost per task (2.5x per token but much lower per task). The cluster crosses ai-frontier, tech-industry, investing and geopolitics with adoption in production by autonomous agents, constituting a tectonic industry movement although not a global black swan.
- Independent sources
- 11
- Cross verticals
- Frontier AI, Tech Industry, Investing, Geopolitics
- Tier 1 actors
- OpenAI, Anthropic, Sam Altman
- Gravity vs. the previous edition
- +1 (84 → 85)
More: Simon Willison, TechCrunch, Latent Space, LessWrong (karma ≥25), OpenAI Blog, Zvi Mowshowitz, Reuters Legal, Techmeme and 2 more
- 09:00
- 14:00
- 02:00Safety overview: GPT-6 Astra · OpenAI Blog
- 14:00
- 14:00Legora reviewed 41 documents in minutes with GPT-6 Astra · OpenAI Blog
- 22:05
- 22:18GPT‑6 Astra · Simon Willison
- 23:09
- 07:18
- 01:59The Pelican comparison grid for Astra is pretty interesting · Simon Willison
- 19:01
- 15:04GPT-6 Astra: The System Card, Alignment and What Comes Next · Zvi Mowshowitz
- 04:31Astra can do a concerning amount with no chain of thought · Alignment Forum
- 01:17Astra is much better at reasoning with filler tokens than previous models · LessWrong (karma ≥25)
- 18:00Cognition helps Devin test its own work with GPT‑6 Astra · OpenAI Blog
- 01:55GPT-6-Astra Can Do Ambitious Things · LessWrong (karma ≥25)
- 02:00Perplexity trusts GPT-6 Astra with end-to-end systems · OpenAI Blog
- 03:39Yet another concerning result on Astra's no-CoT capabilities · LessWrong (karma ≥25)
- 07:47Astra appears to perform belief-propagation-like inference without CoT · LessWrong (karma ≥25)
- 14:00
- 20:00
- 20:03
- 21:00
- 22:31Anthropic and OpenAI release cheaper models as price war intensifies · Financial Times
- 01:46Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war · Simon Willison
- 08:41
- 14:00How invideo improves color grading 3x with GPT‑6 Astra · OpenAI Blog
- 02:00