Which company has best AI model end of August?

Anthropic’s 90.5% price is best understood as a bet on leaderboard persistence. Claude already leads the designated Arena table, and Sonnet 5 gives Anthropic a newly released model capable of defending that position. The hierarchy assigns much lower weight to benchmark announcements, distribution scale, and hypothetical August launches because none directly determines the settlement ranking.

Anthropic combines the current lead with a fresh defensive release

The strongest evidence for Anthropic is unusually direct: it currently holds the leading position on the Arena.ai Text Arena overall leaderboard without style control, the exact table named in the rules. That connection between present performance and settlement explains why Anthropic stands far above Google at 5.7% and OpenAI at 3.4%.

Claude Sonnet 5, launched June 30, adds a second layer. Anthropic says it performs close to Opus 4.8 at a lower cost and has made it the default for Free and Pro users. Wider deployment does not automatically improve an Arena result, but it signals that Anthropic considers the model mature enough for broad use. The inference embedded in the price is that Anthropic can maintain its current ranking through the remaining release cycle, potentially with Sonnet 5 or another model.

The hierarchy discounts benchmarks that do not settle the contract

OpenAI’s GPT-5.6 release supplies the clearest competing technical evidence. OpenAI reported state-of-the-art results of 92.2% on BrowseComp and 62.6% on OSWorld 2.0 for GPT-5.6 Sol. Those scores establish momentum on their respective evaluations. They do not establish first place on the specified Arena table, which is why OpenAI’s probability can remain low despite a recent flagship launch.

Google faces the same translation problem. Google announced Gemini Omni and made Gemini 3.5 Flash generally available across products including AI Studio, Search, Android Studio, and the Gemini app. That reach creates opportunities for feedback and rapid iteration. Settlement still requires a Google-owned model to hold the highest Arena rank at the August 31 check. The evidence that would materially strengthen Google’s case is a model entering the target table near the top and sustaining that position.

Anthropic’s price assumes the leaderboard will remain operationally stable

Several hidden assumptions support the 90.5% outcome. The current leader must remain eligible, retain first place as additional comparisons arrive, and avoid being displaced by an announced or hypothetical unannounced release. The market also assumes that any new model will receive enough Arena evaluation before the deadline to alter the displayed ordering.

Release timing matters because the market closes at August 31, 12:00 AM UTC, while the rules specify checking the table at August 31, 12:00 PM ET. Those timestamps leave roughly a 16-hour interval during daylight-saving time in which the settlement source could change after trading has closed. A leaderboard update, model addition, or ranking recalculation during that window would affect resolution without allowing a corresponding market response.

Specific leaderboard movement would force the largest reassessment

The clearest confirmation for Anthropic would be continued first place after GPT-5.6, Gemini Omni, and Gemini 3.5 Flash have had a credible opportunity to appear and accumulate comparisons. A further Anthropic release could reinforce that outcome; its newsroom documents continued activity, including Fable 5’s global return on July 1.

The strongest weakening evidence would be Google or OpenAI reaching first place on the designated table, especially if the position persisted across subsequent updates. A hypothetical late-August frontier release from any listed company would matter only if Arena added and ranked it before the settlement check. OpenAI’s planned August 26 retirement of o3 from ChatGPT, with GPT-5.4 mini serving as a fallback at rate limits, may influence user perceptions and product routing, though its direct effect on the Arena ranking is unproven.

Fresh rival models remain the main counter-signal

Anthropic’s lead is current, while the deadline is still exposed to active release schedules at three major labs. That is the strongest failure mode for the dominant thesis: a challenger converting recent technical progress into Arena preference quickly enough to take first place.

Market activity offers limited additional confirmation. The event has generated $401,030 in volume and shows $657,660 in liquidity, yet open interest is $44,380. The comparatively smaller outstanding exposure means cumulative volume alone cannot establish durable conviction. The long tail of companies priced between 0.1% and 0.7% primarily represents hypothetical release risk; absent concrete leaderboard movement, the factual contest remains concentrated among Anthropic, Google, and OpenAI.

Sources