Top AI Model Tracker

Kalshi's year-end AI model book holds $5,877,543 of open interest—the deepest single contract in this catalog, deeper than any election, any Fed decision, any championship. Its positions are almost perfectly flat across a price distribution that is anything but.

Claude leads at 70.0% and holds $1,054,530. ChatGPT at 12.7% holds $1,016,743. Grok at 9.8% holds $896,400. Gemini at 7.1% holds $867,740. Muse Spark at 1.2% holds $597,665. Qwen at 0.7% holds $513,788, Ernie at 0.2% holds $465,688, and Kimi at 0.5% holds $464,988.

Eight legs spanning two-tenths of a percent to seventy—a three-hundred-and-fifty-fold spread in price—and the deepest and shallowest positions differ by a factor of 2.3. The book's money is spread as though the race were roughly even; its prices say it is not close.

And today the flow went to the two cheapest legs. Ernie at 0.2% turned $26,086 and Qwen at 0.7% turned $15,020—together $41,106, or 55% of the book's $74,946 day. Claude, at 70%, turned $13,441. The busiest contract in the deepest book on the exchange is the one it prices at one-in-five-hundred.

The silver medal changed hands back. A day ago Kalshi had Grok second at 11.4% and ChatGPT third. ChatGPT is second again at 12.7%, Grok third at 9.8%. Polymarket, which has always disagreed here, now has OpenAI second at 10.0%, Google third at 8.5% and xAI fourth at 6.7%—and prices Anthropic a point and a half higher, at 71.5%.

September has opened and the exchanges are eighteen to one on it. Kalshi's September book prices Claude at 93.0% on $3,951 of volume; Polymarket's prices Anthropic at 93.5% on $70,957. Same question, same answer to within half a point, and eighteen times the flow on the exchange with no open interest at all.

The year-end books, cross-exchange. Kalshi's is the deepest contract in this catalog at $5.88M of open interest; Polymarket's turns $14,374 and orders the field differently below the leader—Anthropic 71.5%, OpenAI 10.0%, Google 8.5%, xAI 6.7%, against Kalshi's ChatGPT 12.7% and Grok 9.8%.

The two ends of Kalshi's year-end distribution: the 70.0% leader holding $1,054,530, and the 1.2% Muse Spark leg holding $597,665—57% as much money on a fifty-eighth of the probability. Read them against each other; that gap is the page.

September, on both exchanges. Kalshi runs company-level and model-checkpoint versions of the same month; Polymarket runs company-level only and turns $70,957 to Kalshi's $3,951—eighteen times the flow on a book with no open interest at all, and the two agree on Claude to within half a point.

Below them, the model-level books. Kalshi's September checkpoint ladder prices five Claude builds and its quoted midpoints sum to 64.5%, not 100; the Arena.ai coding leaderboard beside it has not traded once, and its nine legs sum to 253%.

Breadth, coding, and the Chinese race.

The year-end coding book holds $692,817 and prices Anthropic at 58%, OpenAI at 27%, xAI at 10%—but Google at 1.5% holds $111,538, within $2,448 of xAI's $113,986 at almost seven times the price. The September coding leaderboard has rolled and is quoted but untraded: zero volume, zero open interest, nine legs summing to 253%.

Polymarket's Chinese book turns $42,306 with Alibaba at 74.5%—and its second-busiest contract is Tencent at 1.4%, on $6,824, more than Moonshot took at 14.4%. Kalshi still lists no Chinese model book at all.

The flat open interest against a steep price curve. $1.05M on a 70% leg and $465k on a 0.2% leg is the structural oddity of this page. Either the tail legs are long-dated optionality on a genuine regime change, or they are stale positions from a period when the race looked closer. The distinguishing test is whether tail open interest falls as the year runs out; if it holds into December, it is conviction.

Today the tail did the trading. Ernie at 0.2% and Qwen at 0.7% took 55% of the year-end book's volume between them. That is not stale positioning—it is fresh money going to the legs the book prices at nearly zero, and it is the single most watchable number here.

The silver medal moved back to ChatGPT, and the exchanges still disagree below the leader. Kalshi: ChatGPT 12.7%, Grok 9.8%, Gemini 7.1%. Polymarket: OpenAI 10.0%, Google 8.5%, xAI 6.7%. Google and xAI swap places depending on which exchange you read.

The September checkpoint book does not sum to 100. Five Claude builds quoted at midpoints totalling 64.5%, with claude-opus-5-max marked 24.5% mid against a 43.0% last print. Wide books on thin flow are where the mid and the last stop agreeing, and this one has $8,252 of volume holding it together.

A quoted book with no trades at all. Kalshi's September Arena.ai coding leaderboard shows nine legs summing to 253% on zero volume and zero open interest. It is a clean illustration of what a Categorical book looks like before anyone has taken a side.

Google's $111,538 in the year-end coding book at 1.5%. Second-deepest position in a book where it is priced fourth, and within $2,448 of xAI at seven times the price. Same shape as the headline book, different market.

Whether Kalshi lists a Chinese model book. Polymarket's turns $42,306 a day with Alibaba at 74.5% and Tencent at 1.4% taking more flow than Moonshot at 14.4%. It is the one part of this race with no Kalshi mirror.