In today’s issue:
OpenAI ships GPT-6.2 Astra and Sol
Google launches Gemini 4 Neon
Thinking Machines releases Inkling 2 Mini
936 AI agents publish 936 studies
FrontierHarness tests 22 harnesses
Meta trains memory skills into weights
Alibaba releases Qwen4-Max
Judgment launches Eval Optimizer
Cognition launches Devin Bots
Harnesses help LMs generalize longer
And all the top AI dev news, papers, and tools.
Top Stories
GPT-6.2 Astra and Sol
OpenAI released GPT-6.2 Astra, its new frontier model for the hardest projects, and GPT-6.2 Sol, an efficient model for everyday work. Both reach better results on demanding tasks with fewer output tokens.
Agentic work: On Agents’ Last Exam, which tests multistep professional work, Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost and 5.5 points above Claude Fable 5.2 at 5% of the cost. On GDPval v2, it gains 6.2 points over GPT-6 Astra while using 22% fewer output tokens.
Coding: On Cognition’s FrontierCode 2.0, Sol comes within 0.5 points of Astra at about one-fifth the cost, and OpenAI recommends starting with Sol for day-to-day coding. Astra also sets a new high on Terminal-Bench-Science 0.1, 9 points above GPT-6 Astra.
Pricing: Astra costs $8 per million input tokens, $0.40 cached, and $40 output, 20% below GPT-6 Astra with a 60% cached-input discount. Sol keeps GPT-6.1 Sol’s $2, $0.10, and $10 rates.
Safeguards: Both models meet the Critical cybersecurity threshold under OpenAI’s Preparedness Framework and launch with tighter cyber safeguards. Astra is the first externally released OpenAI model to meet the High threshold for AI Self-improvement.
Availability: Both are live in ChatGPT Work, Codex, and the API, with Ultrafast mode available for both in the API. OpenAI recommends keeping prompt prefixes unchanged and appending new instructions to preserve reasoning across turns.

