Which company has the best AI model on LiveBench (Coding) end of September?

Anthropic’s 60.5% probability is best understood as a wager on benchmark persistence: Claude Fable 5 already leads LiveBench Coding, and the remaining window may favor the incumbent unless OpenAI converts a recent release into a higher score. OpenAI’s 35.5% share prices a specific alternative—GPT-5.6 Sol improves through deployment, iteration, or another eligible release before the September 30 snapshot.

Fable 5’s existing lead gives Anthropic the cleaner path

The latest public LiveBench leaderboard places Anthropic’s Claude Fable 5 ahead of OpenAI’s GPT-5.6 Sol in Coding. That sourced fact explains most of the hierarchy: Anthropic can prevail if the ordering survives, while OpenAI needs the leaderboard to change.

Anthropic also describes Claude Fable 5, restored and rolled out globally on July 1, as its most capable model for ambitious coding projects. A broadly deployed model gives Anthropic an established benchmark candidate and time to refine the surrounding system. The market inference is that Fable 5’s score has enough durability to withstand near-term competition. The 60.5% price still leaves substantial room for displacement, suggesting the current lead alone is insufficient to settle expectations months early.

OpenAI’s probability rests on a release still entering circulation

OpenAI launched GPT-5.6 on July 9 and calls GPT-5.6 Sol its best coding model yet, citing state-of-the-art performance across coding and related tasks. Its July 9 ChatGPT release notes say access is rolling out gradually to eligible plans and may not yet reach every account.

That rollout detail matters because the current leaderboard may capture an early point in Sol’s deployment cycle. A wider release could coincide with configuration changes, improved tool use, or a refreshed submission, although none of those outcomes is confirmed by the supplied evidence. OpenAI’s 35.5% probability therefore represents meaningful release optionality alongside a present benchmark deficit. Evidence that Sol’s score rises after broader availability would strengthen this interpretation. A stable score despite completed rollout would weaken it.

The pricing assumes a two-lab contest through September

Every listed company outside Anthropic and OpenAI sits at 0.7% or below, with most at 0.1%. The implied story is that Meta, Google, Alibaba, DeepSeek, Amazon, Nvidia, and other listed developers are unlikely to introduce an eligible model capable of taking first place by the check date. This is a strong hidden assumption because the contract rewards the single highest Coding score at one specified moment, allowing a late release to matter even without a long public track record.

The available record supplies no comparable upcoming release evidence for those companies, so their low probabilities should be read as market inference, not proof that they lack competitive internal models. A dated model announcement, an official benchmark submission, or a sudden LiveBench entry near the leaders would challenge the two-lab framing immediately.

Resolution mechanics make timing and model ownership decisive

The contract resolves to the company owning the model with the highest LiveBench Coding score when checked September 30, 2026, at 12:00 p.m. ET. This creates a snapshot contest. Average performance across the preceding months, adoption, price, and developer preference do not determine settlement under the stated rule.

A model update arriving shortly before the check can therefore outweigh months of leaderboard stability. The ownership wording also makes the identity attached to the leading model material. Any renamed model, partnership, acquisition, or jointly developed system would require applying the published resolution language to the leaderboard entry.

Score stability is the main counter-signal to an OpenAI comeback

The strongest counterargument to OpenAI’s path is simple: its newest coding model is already public, yet Fable 5 remains ahead in the supplied leaderboard record. If GPT-5.6 Sol completes rollout without narrowing the gap, Anthropic’s persistence thesis gains support. Anthropic could also release an updated Fable variant, extending the target OpenAI must clear.

Concrete repricing catalysts include a LiveBench score refresh, an official model revision from either lab, confirmation that GPT-5.6 Sol’s rollout is complete, or a new entrant reaching the top tier. The market’s $28,080 volume, $21,420 liquidity, and $6,990 open interest support a visible hierarchy, though they provide limited evidence that distant release schedules have been fully incorporated. Until a score or release changes, the current ordering favors the company already leading the settlement benchmark.

Sources