DeepSeek V4-Pro

Nine AI agents fail to meet scientists’ standards across 135 research submissions

The Into the Unknown study found failures in both research quality and integrity. Its automated evaluator also disagreed with scientists on some judgments.

Stanford finds leading AI agents finish under half of simulated healthcare tasks

An October 19 research overview contrasts clinical and administrative benchmarks. Both check completed work, not just plausible answers—and reveal different ways agents fall short.

Audit Finds No SWE-bench Pro Leaderboard Score Has a Complete Public Receipt

Bolt.new Adds Forge, Offering 50x More AI Coding Usage for Opt-In Training Data

The research preview offers a separate experimental lane for software work, but shared prompts, code and repair records cannot be removed once used in training.

Qwen3.8-Max Leads Open-Weight Commerce Agent Bench by Two Passes, With Audit Limits

The benchmark checks whether agents leave commerce software and files in the required state, rather than simply narrate a workflow. But configuration choices and unavailable immutable result bundles limit what the narrow ranking can settle.