AI Daily Newsletter: Mistral Launches Math Problem Solving Model and Anthropic Develops Medicines Autonomously

Ai daily
July 5, 2026

Table of Contents

🧮 Mistral Releases Leanstral 1.5 - A Formal Verification Model, Apache-2.0

Mistral AI has released Leanstral 1.5, a language model specializing in formal verification and proof in Lean 4. The model has 119B parameters in total but only 6B active (MoE), and is fully open-sourced under the Apache-2.0 license. Leanstral 1.5 achieves state-of-the-art results on miniF2F (saturate), solves 587/672 PutnamBench problems, and reaches 87% on FATE-H. Notably, the model discovered 5 previously unknown bugs in 57 open-source codebases tested. Mistral also provides a free API for this model.

Source: Mistral AI Blog

🔬 Anthropic Launches Claude Science and Initiates Drug Development Program

Anthropic has announced Claude Science, a dedicated AI workspace for scientific researchers, integrated with common tools and packages. On the same day, Anthropic declared its own drug development program for “neglected” diseases that Big Pharma considers unprofitable. A researcher at UCSF used Claude Science to detect a viral infection in minutes - something his team had missed for a year. Claude also analyzed 100 rare genetic diseases in under an hour and identified 32 candidates for computational screening. The CEO of Novartis believes AI can shorten drug development time from 12 years to 7-8 years.

Source: The Decoder

⚡ Anthropic in Talks with Samsung for Custom AI Chip Production

According to TechCrunch, Anthropic is discussing the production of a custom AI chip with Samsung, just a week after OpenAI announced its AI chip collaboration with Broadcom. This move indicates that leading AI companies are seeking to become self-sufficient in hardware, reducing their dependence on NVIDIA. Although there is no official agreement yet, Samsung’s advanced chip manufacturing capabilities (foundry) could help Anthropic optimize its own architecture for Claude.

Source: TechCrunch

Thariq Shihipar, a developer at Anthropic, has shared prompting techniques for Claude Fable 5, with the main argument that output quality is now limited by the user’s ability to recognize “unknown unknowns”. He proposes techniques such as goal-oriented brainstorming, structured AI interviews, and detailed implementation notes. An interesting insight: overly specific guidance can lock AI into the wrong approach, while overly open prompts yield generic results.

Source: The Decoder

🏢 Microsoft Establishes AI Deployment Company with $2.5 Billion Commitment

Microsoft has announced the establishment of a new company dedicated to deploying AI for businesses, with a commitment of $2.5 billion. This move follows in the footsteps of Amazon, OpenAI, and Anthropic in having a dedicated deployment team. The new company will focus on bringing Microsoft’s AI solutions (Copilot, Azure AI) into practice for large enterprise customers, including consulting, customization, and operation.