The probability of "claude-opus-4-6-thinking becoming the top AI model on June 13, 2026" has increased from 27.2% to 52.2%.

According to PolyBeats monitoring, in the prediction market Polymarket, in the past 15 minutes, the probability of "Will claude-opus-4-6-thinking become the best AI model on June 13, 2026?" being "No" has increased from 27.2% to 52.2%, with a total buy-side trade volume of $711.

Anthropic released Claude Fable 5 on June 9th, which has been listed on today's LM Arena leaderboard with a score lower than claude-opus-4-6-thinking. Wired and The Wall Street Journal reported on June 11th that Anthropic had originally planned to conduct a "silent downgrade" on accounts suspected of training competitive models using Claude: without notifying users, directly reduce the performance of Fable 5, or weaken the output through prompt modifications, model routing, etc.

Developers and the AI research community criticized this as tantamount to covertly interfering with third-party evaluations and open-source research, especially making it difficult for external security assessment firms to confirm which capability level of the model they are testing.

Subsequently, Anthropic publicly apologized and adjusted its strategy, canceling the secret performance downgrades in such scenarios, and opting for a more transparent prompt or rejection mechanism: if the system detects users building high-capacity AI or training competitive models, it will explicitly reject the request or redirect users to a low-capacity model. This change may increase the public trustworthiness of Fable 5's public evaluations, making the community more likely to believe that it will participate in the leaderboard competition with capabilities close to real-world performance, potentially directly challenging claude-opus-4-6-thinking, currently considered the market leader.
---------------------------------
See the future sooner, follow @PolyBeats_Bot
See tomorrow, today. Follow @PolyBeatsEN