Which company has the best AI model end of September?

Anthropic’s 83.5% price expresses a narrow thesis: its existing Arena lead can be defended by another release before rivals produce an eligible challenger. The hierarchy therefore depends on deployment timing, leaderboard access, and performance at one scheduled observation. Broad claims about which lab has the strongest research program carry limited weight unless they affect that table by September 30.

Anthropic combines the incumbent lead with the clearest dated catalyst

The supplied Arena context places an Anthropic model first on the Text Arena Overall leaderboard without style control. That starting position matters because every rival needs to overtake a visible leader, while Anthropic can prevail through continuity.

Anthropic also launched Claude Opus 4.8 on May 28 and reported improvements across coding, agentic tasks, benchmarks, and professional work. Those company-reported gains help explain the present lead, though the contract ultimately relies on Arena rank instead of Anthropic’s own evaluations.

The stronger causal driver is Anthropic’s announcement that it expects to bring a higher-intelligence “Mythos-class” model to customers in the coming weeks. Market inference: an incumbent with a publicly signaled pre-settlement upgrade has two possible routes to victory—its current model retains first place, or the new model replaces it. The release window gives that pathway more concrete support than an unspecified future launch.

The price assumes Mythos ships broadly enough to count

The announcement contains several hidden assumptions. A Mythos-class model must arrive on schedule, become accessible for Arena evaluation, accumulate enough comparisons to receive a qualifying rank, and perform well under the specific overall, no-style-control setting. A limited preview, delayed rollout, or model absent from Arena would weaken the announcement’s relevance even if Anthropic demonstrates strong results elsewhere.

The rules also assign the outcome to the company that owns the highest-ranked model. That assumes any new entry has a clear identity and ownership mapping. A renamed model, temporary experimental listing, or unresolved affiliation could create rule-interpretation risk. These are hypothetical scenarios; the supplied evidence does not indicate that Anthropic currently faces such a dispute.

Timing adds another dependency. Resolution uses the leaderboard checked at 12:00 PM ET on September 30, while the stated market close is 11:59 PM UTC. The decisive snapshot therefore arrives about eight hours before the listed close. A rollout later that day would miss the specified observation.

OpenAI’s second place prices an unannounced replacement path

OpenAI holds 9.8%, well behind Anthropic yet roughly three times Google’s 3.2%. Its cited release notes say GPT-4.5 was retired from ChatGPT on June 27 and o3 is scheduled for retirement on August 26. Those retirements do not establish weaker Arena performance, but they create a product-cycle gap before settlement.

Market inference: OpenAI’s remaining probability incorporates some expectation of a replacement or upgrade that is absent from the supplied factual record. A named flagship launch, documented Arena availability, or rapid climb toward first place would validate that inference. Continued retirements without a competitive successor would weaken it.

This is also the main counter-signal to Anthropic’s dominance. The market assigns meaningful probability to a company without a cited pre-settlement flagship announcement, suggesting release surprise remains a material part of the distribution.

Google has stronger public evidence than its rank implies

Google says Gemini 3.1 Pro is already rolling out across consumer and developer products, with improved reasoning on complex problem-solving benchmarks. Existing distribution can shorten the path from announcement to evaluation, although Google’s benchmark claims alone cannot determine the Arena result.

A sustained Gemini climb would challenge the assumption that Anthropic’s announced cadence is the best guide to September. Evidence that would matter includes Gemini reaching the top cluster, a further model update becoming Arena-eligible, or Anthropic’s current leader losing rank before Mythos arrives. Without those developments, Google’s rollout remains a plausible catalyst rather than evidence of an impending lead.

Leaderboard movement matters more than the long tail

Meta is priced at 1%, while every other listed company outside Anthropic, OpenAI, and Google sits below 1%. The $198,280 in volume and $479,490 in liquidity show substantial market activity, but open interest of $17,980 provides a narrower measure of currently outstanding exposure. These figures support treating the hierarchy as informative without establishing certainty.

The clearest repricing events are concrete: Mythos ships and enters Arena; its release slips beyond the snapshot; OpenAI announces an eligible successor; Gemini moves into first place; or Arena changes the relevant ranking before resolution. The strongest failure mode for the prevailing thesis is a late rival release that accumulates enough Arena results before noon ET on September 30. Until such evidence arrives, Anthropic’s hierarchy is supported by the combination of incumbency and the most explicit near-term release commitment.

Sources