Question
On 2028-12-31, what will the Artificial Analysis Intelligence Index gap be between the highest-scoring model from OpenAI, Anthropic or Google DeepMind and the highest-scoring model from any AI lab founded after 2023?
Current State and Index Dynamics
The baseline gap today stands at 34 points on Artificial Analysis Intelligence Index v5.0.1, with Anthropic's Claude Fable 5.2 leading at 64.6 and Thinking Machines' Inkling 2 Small as the top eligible post-2023 entrant at 30.6 Artificial Analysis Intelligence Index v5.0.1 Introducing Inkling 2 Small. A critical driver of future variance is the index's re-versioning schedule. The recent shift to the harder v5 (which moved 60% of scoring to private test sets) widened the gap between the overall leader and the open-weights frontier from 3 points to 12 Announcing Artificial Analysis Intelligence Index v5 Artificial Analysis Intelligence Index v5 (major update). By late 2028, further rebasing will likely occur, which tends to push mid-tier models toward the floor and organically inflates point gaps even if underlying capabilities converge.
Neolab Catch-Up Potential
There is a strong case for the gap narrowing over the next two years. Thinking Machines has demonstrated rapid iterative progress and secured a commitment for at least 1 GW of NVIDIA Vera Rubin compute by 2027 Introducing Inkling 2 Small Thinking Machines: Models Intelligence, Performance & Price Thinking Machines Lab inks massive compute deal with .... Safe Superintelligence (SSI) is scaling via an order-of-magnitude compute expansion deal Ilya Sutskever's Safe Superintelligence Inc. and NVIDIA ..., and Discovery Loop combines an elite founding team with Alphabet backing Jeff Dean and other top AI researchers are leaving Google .... If these labs execute well on their gigawatt-scale infrastructure, the most successful entrant should reach the equivalent of today's open-weights tier, which typically lags the absolute frontier by 10 to 20 points.
Incumbent Acceleration and Scale Advantages
However, the catch-up narrative is counterbalanced by incumbent recursive self-improvement (RSI) and structural resource advantages. Anthropic's capability growth is currently 1.88× its historical pace, and METR independently assessed its progress on well-specified tasks at roughly 2×. Because the industry-wide acceleration rate is only about 1.4×, AI-driven R&D gains appear heavily concentrated at the frontier. Incumbents also possess an overwhelming capital moat—Anthropic lists roughly $900 billion in compute commitments, while Alphabet’s 2026 capital spending guidance is around $200 billion. Furthermore, leading labs are already holding back stronger internal models, meaning their true capability ceiling is higher than public leaderboards suggest.
Tails and Key Uncertainties
The median estimate of approximately 23 points reflects partial catch-up by neolabs offset by the incumbents' RSI momentum. Set against related questions, this balance between neolabs' massive compute investments and incumbents' compounding R&D advantages supports a stable distribution, with the central 50% spanning 14 to 34 points. The lower tail (p10 of 7.0) accounts for a surprise breakthrough from SSI or Discovery Loop, or an index redesign that compresses scores at the top. The fat upper tail (p90 of 45.0) captures scenarios where incumbent RSI triggers a compounding feedback loop, harder index versions severely penalize weaker models, or top neolabs pivot away from general models or are acquired by the Big Three.
Set against related questions, this distribution remained stable, balancing the neolabs' massive compute investments against the incumbents' compounding R&D advantages.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited