🤖AI Agents Weekly: Magistral, Agent Bricks, Code Researcher, Automating Workflow Generation, Verified Superintelligence

Magistral, Agent Bricks, Code Researcher, Automating Workflow Generation, Verified Superintelligence

In today’s issue:

• o3-pro is now available via API

• Magistral is Mistral AI’s new reasoning model

• Mistral AI has launched Mistral Compute

• How Anthropic uses Claude Code and builds research agents

• Databricks has launched Agent Bricks

• Microsoft researchers publish a deep research agent for large codebases

• Automating agentic workflow generation

• Morph announces an autoformalization system for verified superintelligence

• Top AI dev news, research papers, and much more.

---

---

Top Stories

---

Mistral Magistral

Mistral AI introduces Magistral, its first reasoning-focused language model, aiming for domain-specific expertise, transparent multi-step logic, and multilingual fluency. Released in two variants, Magistral Small (24B, open-source) and Magistral Medium (enterprise), the model is optimized for real-world applications requiring traceable, interpretable reasoning.

• Magistral Medium achieves 73.6% on AIME2024 (90% with majority voting), while the open-source Small version reaches 70.7% and 83.3% respectively.

• Designed with strong chain-of-thought capabilities across languages, it is well-suited for structured tasks like decision trees, calculations, and rule-based logic.

• The model powers “Think mode” and “Flash Answers” in Mistral’s Le Chat interface, delivering 10x faster response times compared to typical LLMs.

• Magistral is tailored for high-stakes industries (finance, healthcare, law) by ensuring auditability and regulatory compliance through verifiable reasoning chains.

• It's versatile across domains, excelling in legal research, creative writing, software planning, and data engineering, anywhere multi-step reasoning and clarity matter.

Magistral Small is available on Hugging Face under Apache 2.0, while Medium is accessible via Le Chat, API, SageMaker, and upcoming cloud marketplaces. This release is backed by a technical paper.

Paper | Blog