Which Company Has the Best Math AI Model by May?

Anthropic holds a narrow edge in one of the most contested benchmarks in AI right now. The prediction market for best Math AI model by end of May prices Anthropic at 47 percent, a slim lead over a field that includes OpenAI, Google, DeepSeek, xAI, and more than a dozen other contenders. With OpenAI, Google, and DeepSeek all pushing hard on quantitative reasoning, that lead is not comfortable.

The market opened this contract at 36 cents and has climbed to 47 cents as of April 29, 2026. A 9.5-point jump on April 28 reflects real buying pressure, likely tied to Anthropic’s Claude 3.7 Sonnet performance on competition-level math benchmarks. The YES side now prices Anthropic’s win at nearly a coin flip, with NO sitting at 53 percent and representing the combined probability that any other lab takes the crown.

How the Anthropic Math AI Contract Works

This contract resolves YES if Anthropic holds the top-ranked Math AI model as determined by the resolver on May 31, 2026. The resolver evaluates publicly available benchmark data, likely drawing on MATH-500, AIME, and similar competition-math evaluations used across the industry. Anthropic must lead the field outright, not just improve its own score.

• YES (Anthropic wins): $0.47, implying a 47% probability
• NO (any other lab wins): $0.53, implying a 53% probability

The NO side pays out if any single competitor, OpenAI, Google, DeepSeek, xAI, Meta, Mistral, or another lab, posts a stronger math benchmark result by the resolution date. With OpenAI’s o3 and o4-mini already posting top-tier AIME scores, and Google’s Gemini 2.5 Pro competing on the same leaderboards, the NO position reflects a genuinely crowded threat landscape rather than a default hedge.

Sponsored Partner

Market Signals and What the Movement Means

The momentum composite tells a specific story. A flat 1-hour change paired with a 3.5 percent 24-hour gain and a trend score of 29.81 signals fading buying pressure. The contract moved hard on April 28 after Anthropic’s benchmark-related coverage, but the follow-through has stalled. This pattern typically reflects traders booking the news rather than accumulating ahead of a new catalyst.

Volume context matters here. Total traded volume sits at $1,105, with $1,095 of that coming in the last 24 hours. Liquidity is $7,994. This is a thin market. A single large trade can move the price materially in either direction. The current 47 percent figure represents real trader positioning, but it should not be read as a deep, consensus-driven probability.

• Anthropic’s Claude 3.7 Sonnet posted strong results on MATH-500 and select AIME problems in early 2026 benchmarks, driving the late-April price spike.
• OpenAI’s o4-mini set new records on AIME 2024 and AIME 2025 in April 2026, making the math leaderboard genuinely competitive at the top.
• Google’s Gemini 2.5 Pro ranked competitively on Humanity’s Last Exam and similar hard-reasoning tests as of late April 2026.
• The 1-hour change of +0.0% and 24-hour change of +3.5% together indicate the April 28 catalyst is mostly priced in.
• Thin volume means the next benchmark release or model announcement could swing this contract by 10 or more points.

Lines Analysis: Anthropic’s Case and the Crowded Field Against It

Anthropic’s position at 47 percent is defensible but not dominant. Claude 3.7 Sonnet demonstrated strong extended-thinking performance on competition math in early 2026, and Anthropic has signaled that Claude 4 development is progressing. If Anthropic releases a new model or posts updated benchmark scores before May 31, the YES price has room to run toward 60 percent or above.

The threat is real and comes from multiple directions. OpenAI’s o4-mini already leads on AIME under specific benchmark conditions, and OpenAI has a track record of shipping improvements faster than public timelines suggest. Google DeepMind’s AlphaProof lineage gives Google a distinct research advantage on formal math. If either lab posts a new SOTA result on MATH-500 or AIME before the resolution date, the market reprices toward NO quickly.

• Anthropic releasing Claude 4 with published benchmark scores before May 31 would push YES above 60 percent.
• OpenAI updating o4 or o3 with a new AIME record would pressure YES below 40 percent and validate the current NO majority.
• Google posting a Gemini 2.5 Ultra math benchmark in May would split the NO probability further but still cap Anthropic’s ceiling.
• DeepSeek releasing a new math-specialized model would add another credible NO scenario and further dilute Anthropic’s edge.
• The resolution methodology matters: if the resolver uses a single benchmark versus a composite, Anthropic’s ranking could shift significantly based on which test is chosen.

At $1,105 in total volume, this market is priced by a small number of active traders rather than a broad consensus. The 47 percent figure reflects informed positioning on Anthropic’s recent momentum, but the NO side holds majority probability for a reason. The math AI leaderboard changes week to week, and four weeks remain before resolution.

LINES VERDICT

TOO CLOSE TO CALL

Anthropic has earned its place at the top of the market, but a 47 percent probability on a thin-volume contract one month before resolution is a signal of genuine uncertainty, not a lead worth betting heavily on either side.

What the market says: Anthropic sits at 47 percent, just below a coin flip, with the majority of probability distributed across a field that includes OpenAI, Google, and DeepSeek. The thin liquidity means this price is volatile and could shift sharply on any benchmark announcement before May 31, 2026.