Models
Short answer: Models are the cognitive core of Questflow’s financial harness — GPT, Claude, Gemini, and open / finance-specialized models. The model thinks. It is not the alpha source. Skills supply judgment; plugins supply data; execution supplies venues. Questflow is model-agnostic on purpose.
The rule
Agent = Model + HarnessSwap the model, keep the harness. That is how coding agents work and how a finance agent should work. The moat is accumulated context and distilled investor judgment — not which logo is on the API this month.
Model tiers the harness is built to route
Frontier
Hard reasoning, thesis writing, subscriber explanation
GPT, Claude, Gemini
Open / self-hosted
High-frequency scans, pattern jobs, cost control
Llama-class, Hermes-class open weights
Specialized
Sentiment, risk, execution microstructure
Finance-tuned and task-specific models
A router should pick by task: quality vs latency vs cost. Drafting a Fund thesis is not the same job as watching funding every minute.
What models do not do
Invent alpha from a prompt
Replace a fund manager’s invalidation rules
Bypass kill switches and scoped permissions
Excuse a missing skill or missing data plugin
Research Questflow cites (Thinking Machines / Bridgewater, 2026): when everyone has the same public information, edge comes from expert judgment — and models tuned on that judgment beat a raw frontier model on financial tasks. That is why Questflow invests in distillation, not “pick the biggest chatbot.”
How models show up for users
Free / Pro / Max gate AI depth (how much reasoning, monitoring, and auto-execution you unlock), not whether you can trade. Trading access stays open. The model picker is a capability unlock inside the agent workflow, not a separate app.
Related: Four pillars · Coding agent vs finance agent · What is an AI Finance Agent?
Nothing here is financial advice.