Anthropic Ships Claude Fable 5.1, More Than Doubling Its Predecessor on Key Benchmark

🤖 Anthropic just dropped Claude Fable 5.1 and Mythos 5.1 Anthropic says its newest models are its most advanced yet for coding, research, and knowledge work and Fable 5.1 looks like a serious upgrade. Fable 5.1 scores 55.8% on Terminal-Bench 4.0, compared with 42.0% for Fable 5. On Terminal-Bench-Science, it hits 52.6%, more than double its predecessor. It’s not just smarter. It can deliver similar or better results at lower effort levels, while cache reads cost 75% less. Anthropic says this can cut real-world costs by around 25%, and up to 45% for highly agentic workloads. Anthropic is also rolling out stronger enterprise privacy with Enterprise Frontier Safeguards, while claiming its cybersecurity safeguards now flag benign requests about 60% less often. And Fable 5.1 isn’t limited to coding. Its research capabilities are being positioned as an early look at how AI could eventually contribute to scientific discovery. Fable 5.1 is available now. Mythos 5.1, built for cyberdefenders and life scientists, is available through trusted-access programs. Source. @aipost 🏴

🤖 Claude 5.1 Puts Science to the Test Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, the same model with different restrictions. Fable is open to everyone. Mythos requires verification for cybersecurity and life sciences work. With open protein-design tools, almost 50% of designs worked across 12 targets, versus 10-15% normally. Binding affinity was 10 times higher than leading Adaptyv Bio contest entries for 3 targets. The model also built a new elevation map for a third of Venus from Magellan radar images. Resolution improved to 2 to 3 km. Terminal-Bench-Science reached 52.6%, up from 24.7%. Cached reads are 75% cheaper, cutting typical task costs by about 25% and agentic costs by up to 45%. 📊@tech

🙂 OpenAI Has Announced GPT-6 "Welcome to the AGI era," OpenAI president Greg Brockman told reporters while presenting the new model. On the ARC-AGI-3 benchmark, the model scored 98.6%; GPT-5.6 Sol doesn't even reach 8%. So when it comes to picking up unfamiliar situations and abstract reasoning tasks, the model has reached human-level ability, at least within the benchmark. ✈ Employees will no longer have to touch a mouse at all (if they do not want to), OpenAI promises. Until now, developers had to build a separate integration for every application through APIs and connector layers. Astra works with a computer the way a person does: it opens the browser itself, fills in spreadsheets, edits documents, and builds 3D models and dashboards. Hints about where to click and what to type are no longer needed. In the agentic OSWorld 2.0 test, Astra solves 72.6% of tasks correctly, spending 40 minutes on each one. GPT-5.6 Sol scores 65.7% and spends 75 minutes per task. So the "first model of the AGI era" fails more than a quarter of the tasks that are ordinary work for a person. 💰 In price, the model is comparable to Anthropic's Fable line, $10/$50 per 1M input/output tokens. But Brockman