Polymarket traders have pushed the probability of Anthropic releasing its next Mythos-class AI model by September 9, 2026, to 74 percent — a figure backed by real money, real conviction, and a growing market read on how quickly the frontier AI lab moves. Prediction markets are not always right, but they are rarely quiet.

Unlike polls or analyst commentary, every percentage point on the platform maps to a position someone has paid for: contract prices are quoted as probabilities, so the quoted odds reflect where money has actually been staked on each side of the question.

Trading volume on the "next Mythos-class model released by" contract family has crossed $233,000, while the broader AI model race markets on Polymarket have drawn more than $6 million in cumulative volume. That spread in liquidity is worth keeping in mind when reading the headline number — thinner-volume contracts can see their odds move on comparatively few trades, while the larger pools spread positions across many more participants.

What the markets are actually pricing

The September 9 contract sits at 74 percent, but that is just one data point in a cascade of related bets. The odds of a Mythos-class release by September 15 have reached 81 percent, and end-of-month estimates sit at 87 percent.

Beyond the near-term Mythos question, Polymarket traders are also eyeing Anthropic's broader model lineup. A new Opus model arriving by December 31, 2026, carries 96 percent odds, while a new Sonnet model by the same date sits at 85 percent.

The baseline for this optimism is concrete: Anthropic launched Claude Fable 5 and Claude Mythos 5 on June 9, 2026, following a limited preview that began in April. Mythos-class models sit at the top of Anthropic's capability hierarchy, and the June release gave traders a timestamp to anchor their expectations about how long the next cycle would take. Counted from that launch, September 9 falls almost exactly three months later, with the end-of-month deadline just under four months out — the intervals around which this ladder of odds is built.

Why Mythos-class models matter beyond a version number

Mythos-class models are Anthropic's highest-capability tier, distinguished by advanced code reasoning and vulnerability detection at a scale that previous Claude iterations could not reach.

Through Project Glasswing partners, Mythos-class models identified more than 23,000 vulnerabilities. Initial access to these models was limited due to cybersecurity concerns, but it expanded through Project Glasswing, which integrated safety measures for deployment. That expanded access represented a judgment that the safety scaffolding was solid enough to proceed.

What this means for the AI competitive landscape

An 87 percent probability of a Mythos-class model before October implies that traders believe Anthropic is keeping pace with, or ahead of, its own historical schedule. No formal announcement has come from the company confirming a September release, which means the market is pricing in pattern recognition rather than disclosed information.

The 96 percent odds on a new Opus model by year-end add another layer to this picture. Opus sits below Mythos in Anthropic's capability hierarchy but above the mid-tier Sonnet class.

Prediction markets are not oracles. A 74 percent probability still means a 26 percent chance that the September 9 date passes without a release. The deadlines themselves supply the next checkpoints: September 9, September 15, and September 30 each force the near-term contracts to resolve, and the December 31 Opus and Sonnet markets carry the same questions through year-end, turning trader sentiment into a dated, verifiable record either way.