Meta's Muse Code charges $0.20/M tokens, trading deep discounts for training data
shashib · x · 2026-08-10
Meta recently launched the coding agent Muse Code, pricing its output tokens at just $0.20 per million—over 10x cheaper than standard pay-as-you-go rates.
The catch: developers must opt-in to let Meta train on their prompts, completions, and feedback to unlock the discount. In contrast, competitors like Anthropic's Claude Code and OpenAI's Codex do not require this trade-off for competitive pricing.
The author argues this is less about coding agent capability and more about Meta's infrastructure strategy. With Meta raising its 2026 AI capex guidance to $125–145 billion, the company desperately needs visible usage metrics to justify the massive spend to investors. Offering steep discounts effectively buys the user engagement numbers Meta needs, regardless of whether the model closes the capability gap in complex engineering tasks.
Related event: Meta Prices Coding Agent at a Loss to Obtain Training Data(2 posts)→
More from coding & agent
- DIRECT: retrieval-based next-question suggestions for conversational agents — burkov · 2026-10-16
- Claude Code and Codex now autocomplete the user's next action — tokenbender · 2026-10-16
- Hermes Agent ships official Reverse Engineer Anything plugin with one-command setup — Teknium · 2026-10-16
- MLExpert: 35B MoE Agent Trained to Autonomously Handle Machine Learning Engineering — Shangheng Du · 2026-10-16
- Cowork-Gym: Alibaba Synthesizes 64.9K Agent Work Environments, Boosting Qwen by 8.7 Points — Anni Zou · 2026-10-16
- MMSearch-OPD: Joint RL and On-Policy Distillation for Long-Horizon Multimodal Search Agents — PekingUniversity · 2026-10-16