
What happened
AICraft's Auto Router v6 classifies a model: "auto" request into one of 12 task categories, then tries a preset chain of models in order — primary, secondary, tertiary, last resort. For coding, the order is deepseek/deepseek-v4-pro then deepseek/deepseek-v4-flash.
Why it matters
Because the router moves to the next model automatically when one fails, a single auto request can survive an unavailable model without the caller writing fallback branches for each one.
Summaries like this, in your inbox every morning.
Once a team contracts several AI APIs, the decision of which model to call spreads across the codebase, model-name changes break tests, and failover logic goes unwritten. Auto Router v6 is presented as an answer to that sprawl, and the piece is careful to say it is organizing what the documentation states and what response headers reveal rather than pitching the service.
The routing table is copied directly from that documentation, with the caveat that model lineups change and the Model Catalog's IDs should be checked before use. The v6 additions address a subtler issue — that a routing decision made once goes stale — through Lock Decay and Drift Detection, while Fault Self-Healing covers bypassing a failing model for 1 hour. The suggested adoption order is to start with auto, read the headers to see the classification and the model actually used, and pin model IDs only where output must be fixed.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Motive launched Agents, autonomous AI that resolves compliance, maintenance and safety issues without human in…
A solo car-accessories seller had Claude Code build an order-management system and asked it to match sales aga…

The initial Deep Dive design (redundant AI approval stages, a support-judgment step, and document-by-document…

A walkthrough of Data Science Agent in BigQuery Studio used analytics.ec_orders (1,234,567 rows) to define a 9…

Pax8 launched 8os, an operating system for IT service providers to govern AI agents across client businesses…
Anthropic is offering Blackstone a total return of more than 10% to help finance a US$60 billion AI chip procu…
