LLaDA2.2 adds delete and insert to diffusion LLMs for 128K agent workflows
机器之心 · wechat · 2026-07-28
Ant Group’s LLaDA2.2 pushes diffusion language models from static generation into agentic workflows by adding delete and insert to the edit space, not just keep and substitute.
The report describes three big changes: Levenshtein-style editing with four atomic operations, an RL method driven by environment feedback for tool-use and multi-turn trajectories, and native 128K context plus a fixed-capacity block routing design for MoE inference. On benchmarks, LLaDA2.2-flash is roughly on par with comparable autoregressive models on agent tasks, beats them on some tool-use benchmarks, and reaches 1.64x the average BF16 throughput.
Related event: Ant's LLaDA2.2 Brings Diffusion LLMs to Long-Context Agents(2 posts)→
More from coding & agent
- FindDomain lets AI agents check and register domains through an API — geek_guy1 · 2026-07-28
- A talk on AI engineering argues evals, memory, and tools matter more than scattered tool choice — Pavan_Belagatti · 2026-07-28
- Wikipedia MCP server gives LLMs multilingual search and page fetch tools — modelcontextprotocol · 2026-07-28
- Well MCP connector plugs AI into invoices, companies, and contacts — modelcontextprotocol · 2026-07-28
- llama.cpp update breaks older GGUF chat templates in coding-agent workflows — computehungry · 2026-07-28
- Agent builder adds four messaging providers and still can’t reach users reliably — iamstanty · 2026-07-28