Ant Group releases LLaDA2.2-flash, a diffusion model that edits its own output mid-generation
dr_cintas · x · 2026-07-25
Ant Group’s official account shared LLaDA2.2-flash, a diffusion model that can edit its own output mid-generation while keeping block-parallel speed.
- The model is described as a real agent that can plan and call tools.
- Ant says it beats the top autoregressive model on τ²-Bench and MCP-Atlas.
- It reports 1.64× faster throughput and native 128K context.
- The post links to Hugging Face, GitHub, and the official account, suggesting an open release or public artifact is available.
Related event: Ant Group Releases LLaDA2.2-flash(2 posts)→
More from coding & agent
- Hermes pitch claims persistent memory and skill generation can make agents self-improving — Teknium · 2026-07-25
- Cass aims to search across Claude Code, Codex, Cursor and Gemini CLI sessions — doodlestein · 2026-07-25
- GlobalGPT launches a CLI that plugs skills and MCP into Codex — HeyAmit_ · 2026-07-25
- GPT-5.6 and Claude Opus 5 both built a Flappy Bird clone from one prompt — Arindam_1729 · 2026-07-25
- AI agents that read 13F filings often misread puts as bullish long positions — Capedcrusader1923 · 2026-07-25
- MCP’s real business value may be in trust and permissions, not connectors — Warm-Reaction-456 · 2026-07-25