--- title: AI Model Race (Dec 2026) · Octagon Prediction Basket url: https://octagonai.co/prediction-baskets/ai-model-race/ source: Octagon generated_at: 2026-06-11T19:10:49.956Z --- # AI Model Race (Dec 2026) Category: AI / Tech · Horizon: 210 days Themes: AI, Tech, Equities ## Thesis The market is treating Anthropic as the heavy favorite to have the best AI model by end of 2026. We think Google's Gemini is the underpriced contender. ## Overview The Kalshi 'best AI model by December 2026' market has emerged as a real benchmark for tracking who investors think is winning the frontier AI race. Anthropic currently trades at 37¢ on the Yes side — implying the market sees Anthropic as a 63% favorite. Google's Gemini trades at 18¢ — implying just an 18% chance of leading the leaderboard by year-end. We see those probabilities as roughly inverted from where the underlying technical picture is heading. We score frontier-model benchmark deltas every two weeks across reasoning, multimodal, coding, and cost. Anthropic has had a strong recent run but the trajectory is flattening. Gemini, meanwhile, has multimodal advantages baked into the architecture, the DeepMind talent pool, and a TPU-based training stack that gives Google compute advantages competitors don't share. We put Anthropic's probability of leading at year-end closer to 32.5% and Gemini's closer to 37.9%. This basket has two positions, structured as a pair trade. Bet No on Anthropic leading at 37¢ — the larger position because that's where the bigger gap between consensus and our view is. Bet Yes on Gemini leading at 18¢ — the asymmetric long position that pays roughly 5.5-to-1 if Gemini actually leads. We exclude OpenAI from the basket because the contract trades very close to our internal estimate. The two positions express the same underlying view from two sides. If our view is right and the leaderboard shifts toward Gemini through year-end, both positions pay out. If Anthropic just loses ground without Gemini necessarily taking the lead, the Anthropic position still pays. The Gemini position is cheap insurance against the more dramatic outcome. ## Backtest - Since inception (2026-02-21): −21.0% - Live leg coverage: 2 / 2 - Daily candles: 110 _Backtest is hypothetical, computed against the Octagon research report's entry prices. Past performance does not guarantee future results._ ## Legs (Live Kalshi data) | # | Side | Ticker | Market | Allocation | Live YES | 24h Volume | Closes | |---|---|---|---|---|---|---|---| | 01 | Bet No | `KXLLM1-26DEC31-A` | Anthropic (Claude) is not the leading AI model by Dec 2026 | 35% | 69¢ | $15.7k | 2026-12-31 | | 02 | Bet Yes | `KXLLM1-26DEC31-GOOG` | Google (Gemini) leads the AI race by Dec 2026 | 35% | 13¢ | $7.3k | 2026-12-31 | ## How the basket was constructed ### 1. The leaderboard is more open than the market thinks We track 14 frontier-model benchmark deltas every two weeks across reasoning, multimodal, coding, and cost. No single lab dominates across all of those axes. The market has fixated on Anthropic because of recent narrative coverage; the actual benchmark picture is much closer to a four-way race between Anthropic, OpenAI, Google, and a long-tail open-source pack. ### 2. Why Gemini is structurally underpriced Three things favor Gemini through year-end: multimodal advantages that are baked into the architecture (not bolted on), the DeepMind research talent pool, and a TPU-based training stack that gives Google compute economics competitors don't have. None of those advantages have been priced into the 18¢ contract. ### 3. How the pair trade works We hold both positions in roughly equal weight. The Anthropic No leg pays if Anthropic just loses ground — they don't have to lose to Gemini specifically. The Gemini Yes leg pays only if Gemini actually leads. The two together give us a position that profits in the most likely scenarios where our view is right, with cheap upside if the more dramatic version of our view plays out. ### 4. What could go wrong Anthropic ships a major new model in Q3 or Q4 with capabilities that decisively pull ahead on the benchmarks we track. We're watching their model release calendar and any leaks around training compute scale-ups. We're also watching Google's release cadence — if Gemini slips meaningfully on its expected timeline, the long position becomes harder to justify even if the technical setup remains favorable. ## Disclosures Backtested results are hypothetical and do not reflect actual trading. Past performance does not guarantee future results. Octagon is not a registered investment adviser; nothing here is investment advice. Trading prediction-market contracts involves substantial risk of loss. Order routing is to Kalshi; fills are not guaranteed at the prices shown. ## Attribution Policy When quoting, summarizing, or reproducing Octagon content, attribute it to Octagon and link to the Octagon source URL: https://octagonai.co/prediction-baskets/ai-model-race If a specific page was used, cite that page rather than only the site homepage.