🤖 AI Agents Weekly: Claude Sonnet 4.6, Gemini 3.1 Pro, Stripe Minions, Cloudflare Code Mode, Qwen 3.5

2026-02-21 · Source: AI Newsletter · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Cybersecurity & Data Privacy · Depth: Intermediate, quick

Summary

Anthropic released Claude Sonnet 4.6 on February 17, establishing it as the new default model for all Claude users. This update significantly enhances computer use and agentic capabilities, positioning it as a strong coding and agent model within the Sonnet tier. Key improvements include a nearly 5x jump in OSWorld scores from 14.9% to 72.5%, making it highly capable for autonomous computer interaction and GUI-based agent workflows. It also offers a 1M token context window in beta, allowing agents to process extensive codebases and multi-session histories. Users preferred Sonnet 4.6 over Sonnet 4.5 approximately 70% of the time in blind A/B tests, particularly for coding and nuanced reasoning. Priced at $3/$15 per million input/output tokens, it offers cost-efficient scaling for high-volume agent deployments. Additionally, OpenAI and Paradigm introduced EVMBench, a benchmark for evaluating AI agents in detecting, patching, and exploiting smart contract vulnerabilities, revealing agents excel at exploitation but struggle with comprehensive detection and patching.

Key takeaway

For AI Architects and CTOs evaluating agentic models for development or security, Claude Sonnet 4.6 offers compelling advancements in computer interaction and coding, with a 1M token context window for complex tasks. However, be aware that while AI agents show promise in exploiting vulnerabilities, as demonstrated by EVMBench, they currently struggle with comprehensive detection and patching. Prioritize human oversight for critical security auditing until agent capabilities mature further in exhaustive analysis.

Key insights

Claude Sonnet 4.6 significantly advances AI agent capabilities, especially for computer interaction and coding, while EVMBench highlights agent strengths and weaknesses in smart contract security.

Principles

• Larger context windows improve agent performance.
• Agents excel at explicit exploitation tasks.
• Exhaustive auditing remains a challenge for agents.

Method

EVMBench evaluates AI agents on 120 smart contract vulnerabilities from 40 audits, assessing their ability to detect, patch, and exploit high-severity issues.

In practice

• Utilize Sonnet 4.6 for GUI-based agent workflows.
• Consider Sonnet 4.6 for large codebase analysis.
• Benchmark AI agents with EVMBench for security tasks.

Topics

• Claude Sonnet 4.6
• AI Agents
• Large Language Models
• Smart Contract Security
• AI Benchmarking

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, AI Security Engineer

Related on AIssential

• Claude Sonnet 4.6: The Best AI Coding Model Ever! 1M Context, Cheap, & More! (Fully Tested) — Claude Sonnet 4.6 offers Opus-level AI capabilities at half the cost and twice the speed, with a 1M token context.
• Last Week in AI #336 - Sonnet 4.6, Gemini 3.1 Pro, Anthropic vs Pentagon — The AI frontier is rapidly advancing with new model releases, but also faces significant challenges in security and ethical deployment.
• 🗞️ Anthropic Releases Claude Sonnet 4.6, Approaches Opus 4.6 On Many Benchmarks At A Lower Price-point — New AI models offer enhanced performance and cost-efficiency, while security and robotics also see significant advancements.
• Claude Code Just Got A MASSIVE Upgrade! New Agent Skills, Ultra Think, Voice Mode, & More! — Claude Code updates enhance AI agent capabilities through improved skill creation, extended reasoning, and remote access.
• Big upgrade for Sonnet — AI models like Claude Sonnet 4.6 and Gemini's Lyria 3 are rapidly advancing, transforming development workflows and creative tasks.
• Comment remplacer ChatGPT par Claude (et pourquoi) — Claude's agentic AI model, with persistent memory and file interaction, offers a strategic advantage over traditional chatbots.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI Newsletter.