MiniMax M2.7 Debuts as AMD Slams Claude Regression and GPT Image 2 Leaks

Model Releases & Feature Updates

MiniMax M2.7 Model Launch: MiniMax has officially released its M2.7 model, emphasizing high performance and efficiency relative to competitors like Qwen 3.5. Community contributors have already provided various quantized versions, including 1-bit to BF16 GGUF files, to support a wide range of hardware capabilities.

• Minimax M2.7 Released
  • https://huggingface.co/MiniMaxAI/MiniMax-M2.7
• Minimax M2.7 Release Confirmed!
• Unsloth MiniMax M2.7 quants just finished uploading to HF
  • https://huggingface.co/unsloth/MiniMax-M2.7-GGUF
• MiniMaxAI/MiniMax-M2.7 is here!
  • https://huggingface.co/AaryanK/MiniMax-M2.7-GGUF
  • https://huggingface.co/MiniMaxAI/MiniMax-M2.7

Meta Introduces "Contemplating Mode" for Muse Spark: Meta is rolling out a new feature that employs 16 AI agents to collaboratively work on a single prompt to synthesize a more comprehensive and well-researched answer. While the mode aims for higher quality results, users have raised concerns regarding the increased cost and resource consumption required for this multi-agent synthesis.

• Meta started rolling out Contemplating mode for Muse Spark, where 16 agents will work on your prompt to synthesize a consolidated answer!

Industry News & Reliability

AMD Director Reports Significant Regression in Claude Performance: AMD's senior director of AI, Stella Laurenzo, claims that Anthropic’s Claude Code has become less reliable and often fails to complete complex engineering tasks. Citing a chaotic release schedule and frequent outages, Laurenzo confirmed that AMD's engineering team has abandoned the tool in favor of a competing provider.

• AMD's senior director of AI thinks 'Claude has regressed' and that it 'cannot be trusted to perform complex engineering'
  • https://www.pcgamer.com/software/ai/amds-senior-director-of-ai-thinks-claude-has-regressed-and-that-it-cannot-be-trusted-to-perform-complex-engineering
  • https://www.theregister.com/2026/04/06/anthropic_claude_code_dumber_lazier_amd_ai_director
  • https://github.com/anthropics/claude-code/issues/42796

Image Generation

GPT Image 2 Preview Shows Enhanced Realism: A temporary preview of GPT Image 2 has appeared, showcasing significant improvements in photographic detail and realism compared to its predecessors. Though the new model was only briefly accessible before reverting to the original version, the generated samples have generated high interest in the community.

• GPT Image 2 preview

Developer Tools & Integrations

New Mistral AI Skills Pack for OpenClaw and Hermes: A comprehensive skills pack has been released for OpenClaw and Hermes agents, integrating the full range of Mistral AI's features including multimodal capabilities, OCR, and speech processing. The pack is designed for easy integration and includes templates and reference documents to help developers build complex agentic workflows.

• Built a Mistral AI skills pack for OpenClaw and Hermes agent covers pretty much everything
  • http://github.com/DevAgarwal2/mistralai-skills