Polymarket’s AI Model Race: Anthropic Is Still Winning—But the Gap Is Vanishing Fast
Anthropic's Polymarket odds dropped 10 points in a week. OpenAI surged on GPT-5.4-Cyber. Google quietly climbed to 26%. The $3.5M AI model race is wide open.
- Anthropic leads Polymarket‘s “best AI model” market at 54.3%—but dropped nearly 10 points in a week after the Mythos backlash.
- OpenAI surged to 13.5% following the GPT-5.4-Cyber launch, betting that open access beats Anthropic’s whitelist approach.
- Google quietly climbed to 26% as Gemini 3.1 and Gemma 4 gave traders reasons to reconsider the two-horse narrative.
Polymarket traders are having second thoughts about Anthropic—and the timing couldn’t be worse. On the platform’s $3.5 million “Which company has the best AI model by end of June?” market, Anthropic’s odds have dropped from roughly 64% to 54.3% in under seven days, according to Polymarket data. The slide coincides with a brutal stretch: Anthropic released Claude Opus 4.7 on Wednesday, but the goodwill was overshadowed by growing criticism of its Mythos rollout—a 40-org whitelist that security researchers have called elitist.
OpenAI seized the moment. The company launched GPT-5.4-Cyber last week as a direct counter to Mythos, and OpenAI cyber researcher Fouad Matin didn’t bother hiding the competitive framing. “No one should be in the business of picking winners and losers,” he said. The pitch: arm thousands of verified defenders instead of restricting access to a curated club of tech giants. Polymarket traders rewarded the move—OpenAI’s odds jumped from roughly 9% to 13.5%, a 50% gain in market confidence.
Why Google Is the Quiet Winner in the AI Model Race
While Anthropic and OpenAI trade cyber blows, Google has been steadily climbing. The search giant’s Polymarket odds rose from 22.5% to 26%—not flashy, but consistent. Google launched Gemini 3.1 Flash-Lite this week, its fastest Gemini 3 model yet, and released Gemma 4 under an Apache 2.0 license. Neither announcement generated the drama of a Mythos or a GPT-5.4-Cyber, but traders are clearly paying attention. Google also inked a deal with the Department of Defense to deploy Gemini in classified settings—a signal that the enterprise pipeline is real.
The style-control variant of the same Polymarket market tells a different story. When subjective “vibes” are factored in, Anthropic drops to 47.5% and OpenAI jumps to 16%—suggesting that while Anthropic’s models may still benchmark well, the perception war is tilting. Google sits at 22% in style control, its highest reading in months.
There’s a structural problem underneath these odds. Anthropic’s Mythos model remains locked behind a whitelist, which means most of the world can’t actually test it. Claude Opus 4.7 is the safety-tested successor with cyber safeguards dialed back, but it’s not the model that made headlines for finding zero-day vulnerabilities. The market is pricing Anthropic’s potential, not its accessible reality—and that gap is exactly what OpenAI and Google are exploiting.
xAI rounds out the top four at 3.4%, down slightly. DeepSeek sits at 0.9%, up marginally. The rest of the field—Meta, Alibaba, Amazon, Moonshot—collectively holds under 3%. The race is effectively a three-company contest, and right now, the leader is bleeding.
As of Thursday afternoon, $3.5 million in contracts had been traded on the market, with $27,100 in 24-hour volume. The resolution date is June 30, 2026—meaning whoever drops the biggest model upgrade in the next ten weeks takes the pot. Anthropic is still the favorite. But “favorite” and “safe bet” are becoming very different things.
