Anthropic released Claude Sonnet 5 on June 30, 2026 as its most agentic mid-tier model, promising near Opus 4.8 performance at a lower price. The benchmarks look strong, but developers and power users flooded X calling it a flop, pointing to higher token use, heavier refusals, and safety limits. This is the full picture of the backlash.
Sakana AI's new Fugu system says it reaches frontier-level performance by orchestrating several large language models behind one API. The self-reported benchmarks beat Claude Opus 4.8 and GPT-5.5 on several tests, but early users report 30-minute waits, steep pricing, and call it just a router, leaving the orchestrator versus model debate unsettled.
Amazon MGM Studios has dropped Artificial, a nearly finished film about OpenAI chief Sam Altman, just four months after Amazon agreed to invest about 50 billion dollars in OpenAI. The movie, directed by Luca Guadagnino and starring Andrew Garfield, covers the wild week in 2023 when Altman was fired and quickly rehired. Amazon says the film will do better at another studio, and insiders say it paints Altman in a harsh light.
Microsoft unveiled MAI-Thinking-1, its first home‑grown advanced reasoning AI, at Build 2026. The model uses a sparse mixture‑of‑experts design with 35 billion active parameters and can read up to 256 K tokens, about 600 pages. It was trained only on licensed human data, not on other AI models. Tests show it beats rivals in coding and math tasks, and users preferred its answers. Microsoft plans to offer it to businesses for secure, cost‑effective use.
Reach readers who actually finish the story. Advertise on bleepo.
Bots now make up about 57% of all web traffic, beating human visits for the first time, according to Cloudflare data. The rise is driven by fast‑growing AI agents that browse and request pages automatically. Microsoft’s new MAI‑Thinking‑1 model helps companies build even smarter, self‑coordinating AI networks. This shift forces websites to improve bot detection and forces businesses to adopt autonomous AI systems or risk falling behind.
Anthropic’s new AI model Claude Opus 4.7 programmed a four‑legged robot about twenty times faster than the best human‑AI team from the previous year. The model wrote far less code, quickly linked the robot’s cameras and sensors, and built control software that worked on most tasks. It still could not reliably fetch a beach ball, showing real‑time control remains hard, but the speed boost could speed up robot development for many industries.
AWS announced new AI tools at its New York summit to help companies use agents for security, data access, coding and software upkeep. Continuum scans code for flaws and fixes them fast. Context builds a knowledge map so agents can find the right information. Quick lets agents work across apps, while Kiro and DevOps Agent help write and release code safely. Transform keeps software up‑to‑date, and Bedrock AgentCore makes it easier to run agents at scale.
Midjourney, an AI image creator, is spending $74 million to start a whole‑body ultrasound service called Midjourney Medical. The company says its “ultrasonic CT” can scan the entire body in about a minute without radiation or strong magnets, and plans to place 50,000 scanners worldwide by 2031. Radiologists on social media doubt the claims, pointing out ultrasound cannot see through bone or air and that the technology lacks solid medical evidence.
That is every story for now. Check back soon for more.
