GPT-5.5, Claude Opus 4.8 and Gemini 3.5 Pro — the quarter the frontier flattened.
The opening quarter of 2026 delivered the most aggressive run of frontier releases since GPT-4. We rebuilt doja's pipeline mid-flight to use the best of each — and the cost line moved with it.
GPT-5.5 became the cleanest engine for general quality analytics and turning messy input into structured, well-written output. Claude Opus 4.8 set the new bar for coding and design — producing working software and considered interfaces in one pass. Gemini 3.5 Pro put genuinely useful million-token context and deep research into production, making it the obvious choice whenever a task needs to be grounded in real, current sources. None of them won everything. All of them won something — and tighter prompting against the right one cut tokens, retries and refusals in half.
What we changed
- Eureka's build engine moved to Claude Opus 4.8 for coding and design, with GPT-5.5 transforming the founder brief into the spec first.
- Gatsby briefs run on Gemini 3.5 Pro deep research for grounding, then GPT-5.5 for the analytical write-up.
- Limitless's per-stakeholder research is grounded by Gemini 3.5 Pro, message variants are quality-scored by GPT-5.5, and Claude Opus 4.8 designs the sequence itself.
- Every prompt was rewritten to be model-specific — accuracy first, brevity second, cost drops out for free.
The lab that wins next month is irrelevant. The orchestration — and the prompt — that survives every month is the moat.
Tell us what you're building.
Drop us a note — founders, operators, enterprises, press. The team reads every message and replies inside a day.