Smarter Auto Picks + Updated Qwen Token Rates

July 15, 2026

Auto now chooses the least-expensive healthy model that meets your task's quality requirement, then keeps that quality level as your conversation becomes more demanding. Qwen 3.7 Max and Qwen 3.6 Flash also have updated token rates that better reflect their capability.

What you can do

• Let Auto match model quality to the job — simple work starts with a lower-cost capable model; harder requests, deeper analysis, files, and larger tool sets raise the quality floor.
• Keep a capable model once your work escalates — Auto does not step down mid-conversation after it has raised the model quality for your task.
• Use Qwen 3.7 Max for frontier agentic work — its quality score now sits alongside Kimi K2.7 Code, with the same Premium 6× token rate.
• Use Qwen 3.6 Flash for fast, long-context work — it remains the lower-cost Qwen option at the new Light 2× token rate.
• Get a lower Fable 5 token rate — Claude Fable 5 moves from Overkill 24× to Apex 18×.

Where this shows up

• You ask Auto for a quick summary, then turn it into a full analysis with files and tools. Auto can raise quality when the work gets harder and keeps that level for the rest of the conversation.
• You are building or researching across many steps. Auto selects the lowest-cost model that clears the needed quality level instead of spending for more capability than the task needs.
• You want a frontier Qwen model for coding, deep research, or agent workflows. Choose Qwen 3.7 Max when quality matters most, or Qwen 3.6 Flash when speed and long context matter more.

Try it

• "Summarize these meeting notes, then turn the risks into an owner-by-owner action plan."
• "Review this codebase, identify the highest-risk changes, implement the safest fix, and verify it."
• Pick Qwen 3.7 Max in the picker, then: "Research this market, challenge my assumptions, and produce a decision memo with cited sources."

Heads up

• Token rates change for selected models. Qwen 3.7 Max is now Premium 6×; Qwen 3.6 Flash and GPT-5.4 Mini are Light 2×; Claude Fable 5 is now Apex 18×. Your existing plan quota and token balance do not change.
• Auto remains model-agnostic. Pin a model in the picker whenever you need a specific provider or token rate.