Deep|LLM 26Q4: Model Capabilities Advance Rapidly, While Near-Term Monetization Remains Under Pressure

Executive Summary

1. Model capabilities have advanced rapidly over the past three months with the releases of GPT‑6 Astra, Claude Opus 5.5, Gemini 4 Argon and Gemini 4 Carbon. The most important development is their expansion into robotics, video production and scientific research: operating robotic arms, producing videos through code and generating proofs of open mathematical problems (see the section on physical AI, video and science). These physical AI and mathematical capabilities reportedly surprised OpenAI’s own model teams. We view these as emergent abilities, much like coding before them. Further capabilities are likely to emerge as models scale.
2. Near-term commercial returns remain under pressure. Anthropic’s revenue grew 45% quarter over quarter in Q3, down from 144% in Q2, while token spending among businesses in Ramp’s sample barely grew in August and September. Investment continues to rise. The four major tech companies in our sample (Microsoft, Alphabet, Amazon and Meta) are expected to spend about US$730 billion on capex this year, up about US$320 billion from last year. The market expects a further increase of more than US$200 billion next year. Commercial adoption beyond coding is taking longer, leaving a widening gap between what models can do and the revenue they generate.
3. Since mid‑2025, two waves of product adoption have helped revenue catch up with model capabilities. From June to October 2025, the o3 price cut and GPT‑5 launch drove growth in ChatGPT usage. From February to June 2026, Opus 4.5 and 4.6 fueled a surge in coding and enterprise revenue. The next phase of growth will depend on how effectively these capabilities are commercialized.

Key Points