Which company has the best AI Agent end of August?

Anthropic is priced as the incumbent that only has to persist

Anthropic’s 85.5% price expresses a benchmark-persistence thesis. Claude Fable 5 currently ranks first on the live Agent Arena leaderboard, the exact source designated for settlement. That existing lead gives Anthropic a direct evidentiary advantage: its winning condition is continued leadership at the scheduled check, while every competitor needs the table to change before August 31.

This distinction narrows the meaning of “best AI Agent.” The contract does not settle through a broad assessment of commercial adoption, revenue, developer preference, or performance across multiple benchmarks. It uses the highest-ranked model in the Agent Arena table filtered for Models at 12:00 PM ET on August 31. Anthropic’s price therefore depends on one observable ranking surviving until one specified snapshot.

Fable 5’s product design supports the leaderboard story

Anthropic announced Claude Fable 5 on June 9 and described it as built for long-running agentic work. That stated focus fits the category being measured and provides a causal explanation for its current first-place position. The market can connect a product designed for agentic tasks with evidence that it already leads the designated agent leaderboard.

Availability also matters because leaderboard standing may depend on models remaining accessible for evaluation. Anthropic said Fable 5’s global availability was restored on July 1 after a temporary export-control interruption. The restoration supports continued participation through the resolution window. The earlier disruption identifies a failure mode: another access, policy, or deployment interruption could affect usage or sampling before the final check. The supplied record does not establish that such a disruption is expected.

OpenAI’s 13% price concentrates the release-cycle challenge

OpenAI is the only competitor carrying a double-digit price, at 13%. The documented catalyst is GPT-5.6, officially launched July 9 across ChatGPT, Codex, and the OpenAI API. Broad distribution gives the model several channels through which users and evaluators can encounter it, creating a plausible path to an Agent Arena ranking change before settlement.

That path remains conditional. A launch announcement and platform availability do not establish a future leaderboard result. For OpenAI’s case to strengthen, GPT-5.6 would need to appear in the relevant Agent Arena table and post results strong enough to overtake Fable 5 by the specified check. Evidence that GPT-5.6 remains below Fable 5 after meaningful leaderboard exposure would weaken the release-driven thesis, especially as August advances.

The hierarchy assumes the current ranking is durable and representative

Several hidden assumptions sit beneath Anthropic’s lead. The first is that Agent Arena’s model set and results will remain sufficiently stable. The second is that Fable 5’s current rank survives new evaluations, model updates, and fresh entries. The third is operational: both the leaderboard and Anthropic’s model remain available when the table is checked.

The remaining companies are each priced at 1.4% or below, led by Moonshot at 1.4% and Alibaba at 0.6%. The supplied evidence contains no corresponding announcement or current first-place result for those names. Their prices therefore represent low-weight hypothetical paths such as an unannounced model release, a sudden leaderboard entry, or a sharp ranking reversal. A verified move into first place would supply the missing evidence and force a reassessment of the two-company hierarchy.

The final leaderboard check matters more than late August headlines

The clearest repricing catalysts are concrete table changes: GPT-5.6 entering near the top, Fable 5 losing first place, a new model taking the lead, or a material access change affecting either frontrunner. Official updates from Anthropic or OpenAI would matter chiefly through their subsequent effect on the designated leaderboard.

Timing creates an additional constraint. Resolution uses the leaderboard at 12:00 PM ET on August 31, while the market’s listed close is 11:59 PM UTC that day. The decisive observation therefore occurs several hours before the stated close. Releases or ranking changes after the prescribed check should have no bearing on settlement under the written criteria.

The $62,760 in volume and $46,730 in liquidity show meaningful activity around this hierarchy, while $9,030 in open interest is smaller than either figure. Those totals document engagement; they do not validate the assumption that today’s leader will persist. The strongest counter-signal is still GPT-5.6’s recent rollout. The strongest evidence against that challenge would be continued Fable 5 leadership after GPT-5.6 has had a clear opportunity to register on the settlement leaderboard.

Sources