Which company has the best AI model end of September?

Anthropic’s large lead rests on a timing advantage: it has recently launched frontier models while already holding the top arena.ai Text Arena positions. That makes the contest chiefly a question of leaderboard persistence through September 30, not a broad referendum on which lab has the strongest research program, largest user base, or most ambitious product roadmap. The market assigns Anthropic 87.5%, with OpenAI at 9.1% and Google at 2.3%, signaling that a meaningful reversal would require a rapid change in the ranking that governs settlement.

The decisive asset is an established leaderboard position

The supplied research states that Anthropic held the top arena.ai text leaderboard spots as of September 2, 2026. This is the most consequential fact because the contract resolves according to the company owning the model with the highest rank in arena.ai’s Text Arena Overall table when checked on September 30 at noon Eastern time. A company does not need to win a long-term model-development race to win this event; it needs one qualifying model at the table’s summit at the specified checkpoint.

Anthropic also announced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, calling them its most advanced models for coding and knowledge work. That release sequence provides a plausible explanation for the present hierarchy. New models can help defend an existing lead if their performance is incorporated into the relevant Arena ranking before the cutoff. The market’s concentration around Anthropic therefore embeds an assumption that its leading models will retain sufficient comparative strength and that no rival’s newer release will achieve a higher Overall rank in time.

OpenAI’s case depends on GPT-6 Astra translating quickly

OpenAI has the clearest near-term path to a material repricing because GPT-6 Astra was announced on September 3, 2026. OpenAI described Astra as its most intelligent and aligned model yet. Its 9.1% market probability, combined with a 0.5 percentage-point gain over the prior 24 hours, places OpenAI well ahead of every challenger outside Anthropic.

The key hidden assumption is that an announcement of a major model does not automatically establish a higher Arena rank. The settlement criterion concerns arena.ai’s Overall table, making the relevant evidence the model’s recorded position there. A hypothetical catalyst for OpenAI would be GPT-6 Astra appearing on the applicable leaderboard and moving above Anthropic’s highest-ranked qualifying model. Confirmation would be especially strong if that lead persisted as the September 30 check approached. Conversely, Astra’s absence from the relevant table, or a ranking below Anthropic’s leading entry, would weaken the case that its announcement alone can alter the outcome.

Google needs a larger ranking jump than its release headlines imply

Google announced Gemini 3.8 Flash and Gemini 3.8 Flash Cyber on September 2, describing Gemini 3.8 Flash as its best reasoning and coding model yet. Those claims establish a concrete product catalyst, particularly because reasoning and coding performance can influence comparative model evaluations. Yet Google’s 2.3% probability indicates that the present market inference assigns limited odds to Gemini converting its launch into the single highest Overall rank before the deadline.

Google’s route is narrower than Anthropic’s because it begins from outside the reported leading positions. Its route is also narrower than OpenAI’s in market terms, given the gap between their quoted probabilities. Evidence that would challenge this ordering is straightforward: Gemini 3.8 Flash taking a leading Arena position, followed by a clear placement above the best Anthropic and OpenAI entries at the relevant time. Evidence that would preserve the current ordering would include Anthropic retaining the highest-listed model despite Gemini’s release.

The deadline turns model rollout cadence into the central risk

The September 30 measurement date and October 1 close create a short window in which release timing, leaderboard inclusion, and rank movement carry unusual weight. A lab can announce a model close to the checkpoint, yet the event’s outcome still depends on the table at the prescribed moment. This favors the incumbent insofar as its models have already established the reported top positions, while giving recently released challengers a limited period to demonstrate a higher rank.

The $2.88 million in volume, $1.61 million in liquidity, and $470,690.69 in open interest indicate substantial attention to this narrow ranking question. Anthropic’s probability fell one percentage point over 24 hours while OpenAI rose 0.5 points. That movement is consistent with the inference that GPT-6 Astra’s launch has introduced a live competitive threat, although the supplied record does not establish a corresponding Arena ranking change.

The main counter-signal is that one leaderboard update can outweigh the lead

The strongest challenge to Anthropic’s position is structural: the resolution rule selects one company based on the highest model rank, leaving little value in a broad portfolio of capable models if a rival holds the top entry at the checkpoint. OpenAI’s Astra and Google’s Gemini 3.8 Flash are both fresh named releases from major competitors, providing identifiable routes for that outcome.

For Anthropic, the confirming evidence is continued leadership in the arena.ai Text Arena Overall table through September. The principal weakening evidence is any verified overtake by GPT-6 Astra or Gemini 3.8 Flash before the September 30 check. Other listed companies carry very small probabilities, so the supplied market context assigns their potential leap to first place a limited role in the current causal story.

Sources