# anthropic

Opus 4.5 went from 42% to 95% on CORE-Bench without any change to the model. What changed was a broken test and an overly restrictive harness. The part of Anthropic's agent evals guide most worth reading is its string of cases where the score lied, plus the advice it keeps coming back to: read the transcripts before you trust the number.

Anthropic Fellows adapted software diffing for AI safety, creating a cross-architecture tool that compares model behavior. It found a “Chinese Communist Party Alignment” switch in Chinese models and an “American Exceptionalism” switch in an American model.

Anthropic has revealed how its Managed Agents architecture evolved: by separating sessions, harnesses, and sandboxes, it went from “pets” to “cattle.” The result was a reduction of more than 90% in p95 time to first token, while also solving problems with security and scalability.

Claude Code shipped three releases in three days in early April (v2.1.94 → v2.1.101), spanning security hardening, new commands, and performance gains. From command injection fixes to subagents finally getting names, the release cadence feels like the whole team downed ten shots of espresso.

Anthropic has finally released an official CLI tool, ant. Written in Go, it features a resource-oriented architecture, built-in file injection, and GJSON transforms. No more manually assembling curl headers every time you call the Claude API.

Anthropic’s annualized revenue has topped $30B, passing OpenAI’s $24B, while its model-training spend is about one-quarter as large. SaaStr breaks down six angles showing how the AI arms race’s business logic is shifting.

AMD AI director Stella Laurenzo analyzed nearly 7,000 Claude Code sessions and found that after the February update, thinking depth fell 67%, the Read:Edit ratio dropped from 6.6 to 2.0, and estimated daily API costs rose from $12 to $1,504. Claude Opus wrote the report from its own logs.

Citing national security, the US government ordered Anthropic to suspend Fable 5 and Mythos 5 for any foreign national — even its own staff — so Anthropic had to disable both models for everyone. The alleged 'jailbreak'? Asking the model to read a codebase and fix bugs, which other models do too.

Clawd.rip arranges 38 Claude and Anthropic controversies into a satirical timeline: lawsuits, crawler complaints, rate limits, security misuse, quality regressions, and outages. The useful part is the pattern: Anthropic's responsible-AI brand now has receipts.

Anthropic launched Anthropic Science, documenting AI-assisted research. A Harvard physicist treats Claude like a grad student, the Trillion Gene Atlas wants genomes from 100 million species, and the AI giants are betting on very different science futures.

Anthropic's third Economic Index finds broader use cases, more lower-wage tasks, and a four-point success-rate edge for experienced Claude users. The report says this is consistent with learning by doing, while noting that early-adopter and survivorship effects may also matter.

Anthropic launches Claude Code Channels with native Telegram and Discord support, turning Claude Code into a 24/7 always-on AI agent. VentureBeat calls it the OpenClaw killer.