State of AI for development: November 2025

Three things from November that we think change decisions, and what we are telling clients to do about each.

The week of state-of-the-art coding models

Three frontier labs claimed the top coding benchmark within one week: Gemini 3 Pro, then GPT-5.1 Codex, then Opus 4.5. Every frontier model now performs well at coding, with real differences of character rather than capability (Gemini feels substantially the best for UX work). The tight spread suggests the labs have superior versions still to come.

What we are telling clients: stop tracking a single leader. The top is now a moving tie, so pick models per task and expect the pace of improvement to hold rather than slow.

Frontier intelligence became affordable for daily work

Opus 4.5 launched at 30 per cent of the cost of its predecessor, and in use it tends to spend fewer output tokens on the same work because it works smarter (it looks up how-tos before taking on larger tasks). It is now the default model in paid Claude Code at the same usage caps as the mid-tier model.

What we are telling clients: make it the daily driver if you are in Claude Code, and re-run any cost calculation you did on the old pricing. The cost objection to using frontier intelligence for everyday development is largely gone.

Multi-model orchestration is becoming a practice

More teams are splitting coding work across models: one for planning, one for analysis, one for execution. The tooling for this is still rough, but the shape is clear and the model releases above make the split more attractive, since different models now lead at different roles.

What we are telling clients: know your preferred model for each role, write the choices down, and revisit them monthly. The picks age fast; the habit of having picks does not.

---

This digest is the public edition. Advisory clients get the full monthly review, a tailored edition for their context and the conversation that goes with it.