Mark RadarMARK RADAR
About
中 EN
Sign in
Event File AI Open-Source AI Models DeepSeek

Meta Expands Muse Code as DeepSeek, Z.ai Push Agent Models

1 reports · First detected 2026-09-01 · Last active 2026-09-01

AI coding tools are evolving from chat-based assistants into systems that can plan, edit and validate changes across large software repositories. That shift has made developer workflows a new battleground for Meta, DeepSeek and Z.ai, with competition extending beyond model size to multi-agent coordination, open-weight availability and inference costs. The outcome could shape enterprise adoption as companies weigh productivity gains against data-governance risks involving proprietary code.

Meta took Muse Code out of beta on Aug. 31, adding a developer-preview SDK, cross-session messaging and multi-agent workflows. Monthly plans were introduced at $5, $15 and $50. DeepSeek also released the 168-gigabyte, MIT-licensed weights for its experimental DeepSeek-V4-Flash-Vision-Exp model. Z.ai, meanwhile, unveiled GLM-5.3-Flash on Aug. 30, reporting scores of 63.4 on DeepSWE v1.1 and 48.8 on AutomationBench v1.0.6.

All Coverage

1 original reports
NEWS.SMOL.AI 2026-08-31
not much happened today

The Backstory

The history behind this event
Meta Launches Muse Code to Challenge OpenAI, Anthropic2026-08-06 · 7 reports · similarity 0.87

Meta Platforms is expanding its generative-AI push into software development, taking aim at terminal-based coding tools led by Anthropic’s Claude Code and OpenAI Codex. Meta Superintelligence Labs introduced the Muse Spark model family on April 8 and opened a paid developer API on July 9. Muse Code marks a broader effort to turn Meta’s AI research into commercial tools for engineers handling complex, repository-scale work.

Meta AI released Muse Code in beta on Aug. 5, powered by the new Muse Spark 1.2 model and available on macOS and Linux. The agent can plan changes, write code, run tests and verify results, while persistent asynchronous background agents continue working across long sessions. Standard API pricing is $1.25 per million input tokens and $4.25 per million output tokens. Meta said one demonstration ran for up to 24 hours and made more than 1,000 tool calls.

Mark Radar｜MARK RADAR
Past Issues・by MarkReadFintech

If you search news on Google, you can set Mark Radar as a preferred source—our coverage will show up more often in your results. Set as preferred source on Google →

All times are in Taipei time (GMT+8)