The rules behind AI access, pricing and code review

GitHub ties Claude Fable 5.2 to a retention decision, Meta prices long Muse Ember prompts separately, and AutoRefine tests what AI code judges can learn from workflow outcomes.

3 stories · Yesterday in AI

TL;DR

The short version
  1. Copilot Business and Enterprise admins have to enable Fable 5.2 and confirm how Anthropic will retain prompts and outputs.
  2. AutoRefine improved agreement with current code review outcomes on a small test, while the previous judge did better on older work.
  3. Muse Ember’s listed input and output rates triple above 200,000 input tokens, a threshold that matters for repeated document work.

Enterprise Ops

1 story

AutoRefine tests what code judges can learn from merges

Meta researchers improved an AI judge’s agreement with recent merge and abandonment outcomes in a small test, while the older judge still did better on earlier work.

As a leader: Measure both missed useful changes and passed weak changes before making a recalibrated judge a release gate.

As a practitioner: Record why a change was abandoned, and compare that reason with the information the judge actually receives.

arXiv / Shervin Shahidi, James Poon and colleagues ↗Research preprint

Vendors

2 stories

Claude Fable 5.2 gets a policy switch in GitHub Copilot

GitHub leaves Claude Fable 5.2 off by default for Copilot Business and Enterprise, where enabling it acknowledges Anthropic’s default safety-monitoring retention.

As a leader: Decide which repositories can use Fable 5.2 after confirming the retention terms and exception for your Copilot account.

As a practitioner: Check the model picker against your repository’s data policy before sending sensitive code to Fable 5.2.

Anthropic ↗Vendor documentation

Meta’s Muse Ember charges a higher rate for long prompts

Meta launched Muse Ember for repeated agent tasks, with input and output rates that triple when a prompt exceeds 200,000 tokens.

As a leader: Model cost per accepted result using the actual mix of short prompts, long prompts and escalations.

As a practitioner: Run a sample of routine extractions and long-document requests, then compare token use and review effort by prompt size.

Meta AI Research ↗Vendor documentation