#coding agents

Cerebras released MiniMax-M2-REAP-162B-A10B, a REAP-pruned SMoE checkpoint that trims experts to reduce memory while keeping near-original performance on code, reasoning and agentic tool calls.

'GLM-4.6 expands context to 200K tokens, reduces token consumption on real-world coding benchmarks, and offers open weights for local inference and research.'

'Discover how agentic AI in 2025 is reshaping workflows with reasoning-driven RAG, conversational voice agents, coordination protocols, DeepResearch systems, and autonomous coding and CUA tools.'