Google's Gemini 3 Pro Preview has achieved a 37.6% accuracy score on the FrontierMath benchmark, setting a new industry record for expert-level mathematical reasoning. With Polymarket traders assigning a 37% probability to this outcome being confirmed by January 31, 2026, investors are watching closely as Google's AI leadership could significantly impact Alphabet stock performance.
Current Situation
Alphabet (GOOG) shares are trading near $316 in January 2026, with analyst price targets ranging from $326 to $400. The stock has shown renewed strength following Google's AI progress, particularly the Gemini 3 launch. Market sentiment has shifted favorably as OpenAI's GPT-5 missteps allowed Gemini to capture market share, with Google now holding 20% of the enterprise LLM market.
Benchmark Performance Analysis
| Metric | Gemini 3 Pro Preview | GPT-5 (high) | GPT-5.1 (high) |
|---|---|---|---|
| Tier 1-3 Accuracy | 37.6% | 32.4% | 31.0% |
| Problems Solved | 109/290 | - | - |
| Tier 4 (Research) | 18.8% | 12.5% | 12.5% |
| Deep Think Mode | >40% | - | - |
Technical Context
FrontierMath is a benchmark comprising 350 original mathematics problems created by expert mathematicians, covering number theory, algebraic geometry, category theory, and real analysis. Solving typical problems requires hours of specialist effort, with Tier 4 problems demanding days of work. The unpublished nature of problems prevents training contamination, making scores highly reliable indicators of genuine reasoning capability.
Key Factors
Google DeepMind's historical strength in mathematical AI, evidenced by projects like AlphaGeometry and AlphaProof, has positioned Gemini 3 for benchmark dominance. The 37.6% score represents a substantial lead over OpenAI's best models, with the gap widening to 6 percentage points on the most difficult Tier 4 problems.
Analyst confidence in Alphabet remains high, with firms including Jefferies, Melius, and Pivotal lifting price targets toward $365-$400. Google's TPU innovation, supported by nearly 400 patents and major client deals including Anthropic's one million TPU agreement, suggests sustained competitive advantages in AI infrastructure.
The Polymarket probability of 37% appears to reflect uncertainty around verification criteria rather than benchmark achievement itself, as the 37.6% score has already been published by independent evaluators.
Prediction
Direction: Bullish Probability: 72% Horizon: 11 days (January 31, 2026) Answer: Yes
Gemini 3 Pro Preview has already achieved 37.6% accuracy on FrontierMath Tier 1-3, exceeding the 37% threshold specified in the prediction market. The verification by Epoch AI and publication on official leaderboards suggests the outcome will be confirmed by the January 31 deadline. Market underpricing at 37% probability likely reflects information asymmetry rather than genuine uncertainty about benchmark performance.
Sources
Technical Analysis
365 trading days of data for GOOG (2024-08-05 to 2026-01-16)