AI Newsletter

Polymarket rates Anthropic with 88% probability as the provider of the best AI model by end of September 2026 – with $2.2M trading volume. Claude Fable 5.1 and new security features like OS sandboxing (Seatbelt/bubblewrap) solidify Anthropic's technical leadership. OpenAI ranks at only 9% with GPT-6 Astra – a dramatic lag that reflects the competitive shift in the frontier segment.

Polymarket lists the probability of an official OpenAI AGI announcement before 2027 at 28% – with strong increase of +19% in the current month. Sam Altman indicated internally that models beyond GPT-6 Astra are already in development and could achieve genuine scientific progress. The combination of rising market bets and Altman's statements suggests a strategic escalation in AGI communication.

Jacob Coxon leaves Anthropic under protest, accusing the company of using AI safety regulation to exclude competitors from the market. The allegation: Anthropic and OpenAI coordinate lobbying efforts that effectively function as market entry barriers for smaller providers. The resignation is highly political – it hits Anthropic precisely during its IPO process and at peak market valuation.

Mistral positions itself as inference infrastructure for Chinese models like GLM 5.2, effectively surrendering its claim as an independent European frontier lab. Critics view this as strategic capitulation: Mistral's Large 3 flagship is reportedly already based on Chinese architectures. For European AI sovereignty, this is an alarm signal – the EU champion becomes a distribution channel for Chinese technology.

Andrew Tulloch, to whom Meta offered a compensation package of over one billion dollars, is leaving the company. The departure signals that even astronomical salary packages are no guarantee for talent retention in the frontier AI segment. For Meta, the loss represents a significant setback in the competition for top talent – amid the most expensive AI talent war in history.