AI Dev Weekly #31: Claude Duet, Beam, Mercury 3 and longer prompt caching

Anthropic launched Claude Duet, a proactive agent on Opus 5.5, as a US-only research preview for Pro and Max subscribers. Reflection released Beam weights (501.3B parameters, Apache 2.0) and Meta published Muse Spark Open (816B), while Inception's diffusion-based Mercury 3 offers a 1M-token context at $0.20/$0.70 per million tokens.

DEV Community

Open source

AI Dev Weekly #31: Claude Duet, Beam, Mercury 3 and longer prompt caching

Claude Duet runs on Opus 5.5 and keeps working on a VM with its own browser while devices are off
Beam: 501.3B total parameters, 23.4B active, Apache 2.0 license
Muse Spark Open: 816B total parameters, 42B active, Muse Model License
Mercury 3: 1,068 tokens/s on Nvidia GPUs, $0.20 input and $0.70 output per million tokens

#Anthropic#Claude#Meta#Inception

AI Dev Weekly #31: Claude Duet, Beam, Mercury 3 and longer prompt caching

Inception releases Mercury 3, its flagship diffusion LLM

Inception Labs launches Mercury 2.5, a diffusion LLM at 1,107 tokens/sec

OpenAI adds 12-hour prompt cache to Responses API

Experiment measures how long Claude Code's prompt cache lasts