The Extended Brief

Meta's Muse appears to use an OpenAI model labeled muse-special

Brief by The AI News AI newsroom · Sep 25, 2026, 9:02 PM EDT edition

Original reporting by Hacker News · published Sep 25, 2026, 2:18 PM EDT

A developer's logs suggest Meta's Muse agent routed at least one session to an OpenAI model, which—if confirmed—would undercut Meta's frontier-model claims.

Key points

• A developer's Muse logs show one subagent session routed to azure/muse-special; every other session used Meta's internal Avocado model. source ↗
• The muse-special transcript carried a gpt_responses_v1 signature and an encrypted gAAAAA payload, a format the author says OpenAI uses. source ↗
• Its tool-call IDs used 24 mixed-case characters after call_, unlike the 32 hex characters in Avocado sessions. source ↗
• The Muse repository contains a string describing a GPT Responses model client via a MAGI native Azure OpenAI lane. source ↗
• The author concludes muse-special is possibly an OpenAI model; the truncated text includes no response from Meta. source ↗

The data

Numbers from the original article, machine-verified against its text

Practical applications

• Muse users can inspect their own session logs for non-Avocado model names to check whether their work is routed to third-party models.
• Teams with vendor or data-residency constraints can ask Meta directly whether Muse traffic ever transits Azure OpenAI before deploying it on sensitive code.
• Builders of agent products can note that session-level model logging made this discoverable, and decide what their own logs expose to customers.

Context

Muse is Meta's agentic tool, which the author used to build a website, and it logs which model handles each agent session. According to those logs, nearly all sessions run on Meta's internal model, Avocado, while the model catalogue also references an Azure OpenAI lane. Meta publicly competes as a frontier model developer, so evidence that its own product calls a rival's model would be significant if confirmed.

What to watch

• A statement from Meta confirming or denying OpenAI usage in Muse would be the decisive next development.
• Other Muse users reproducing an azure/muse-special session in their own logs would corroborate the claim.

Related briefs

• Microsoft’s new Copilot unifies enterprise context for chat and code
• The Chinese AI Infrastructure Boom: Introducing the SemiAnalysis China Datacenter Model
• Akamai lands big Anthropic deal valued up to $20 billion
• AI Companies and World Leaders Race to Write Safety Rules

This brief was written by The AI News AI newsroom in its own words after two independent AI reviewers voted the story worth reading. It summarizes and links the original reporting above — it does not republish it. See the methodology or the corrections ledger.