Reflection’s first model, Beam, scores 30 on the Artificial Analysis Intelligence Index, placing it #2 among US open weights models Beam is an open weights model with 501B total and 23B active parameters, released on @huggingface Congratulations to @reflection_ai on the release! Key takeaways: ➤ Second among US open weights models: Beam scores 30, behind only Meta's Muse Spark Open (39) and ahead of yesterday’s NVIDIA Nemotron 3.5 Super VL (26) release. Outside China, only Muse Spark Open and the UAE's K2 Horizon (31) score higher ➤ Token efficient for its intelligence: Beam uses ~26k output tokens per Intelligence Index task compared to K2 Horizon, one point ahead, which uses twice as many (~52k). Proprietary models still set the frontier, with GPT-6 Astra (max) scoring 53 on the same budget (~27k) ➤ Strong on agentic workflows, weaker in the terminal: Beam scores 48.5% on AutomationBench-AA, 11 percentage points ahead of K2 Horizon (37.2%) and level with Qwen3.8 27B (48.2%). Terminal-Bench 4.0 is a weak point, where Beam scores 3.0%, behind Qwen3.8 27B (5.6%) and Muse Spark Open (8.1%) ➤ Knows more, guesses more: On AA-Omniscience it answers 33% of questions correctly, well ahead of K2 Horizon (18%) and Qwen3.8 27B (16%). But its hallucination rate is 47%, against 26% for K2 Horizon and 30% for Qwen3.8 27B, giving Beam a net index score of 1.5, ahead of both peers (−3.0 and −10.0) Other model details Context window: 1M tokens Modalities: Text Parameters: 501B / 23B

Armenia can build AI companies with global reach. Computing infrastructure, talent, and entrepreneurship need to grow together. I’ll join Friends of Yerevan State University in Burbank on October 7 to discuss what it takes. 🇦🇲 Event details: linkedin.com/feed/update/urn…

Happy to share that AIRA₂ has been accepted at NeurIPS 2026! 🎉 Sequel to AIRA₁ (NeurIPS 2025 Spotlight), and predecessor to AIRA₃ -- which just won its first live Kaggle gold. Great to see this work compound.

TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission? medium.com/@TmlrOrg/asking-a…

AI has crossed an important threshold in job readiness. We believe AI agents will soon be able to serve as an independent workforce across industries, likely within the next one or two years. We see this as a generational opportunity to make human work more valuable and human lives richer. Humans and AI have complementary strengths. Humans are not biologically best equipped for many parts of modern work. It strains our working memory, exhausts our attention, and demands a precision we struggle to sustain. We did not evolve to fill out tax forms or sift through decades of records. We had to do this work because it required a level of intelligence that, until recently, only humans could provide. With AI sharing that burden, we can focus more of our cognitive power on the things that truly matter: understanding people, asking the right questions, exercising judgment and taste, and taking responsibility for the outcomes. We believe in a future where human–AI teams become 10–100 times as productive as individuals working alone today. That future will take deliberate effort. Researchers need to understand and improve how agents learn. Builders need to turn those capabilities into systems people can trust. Organizations need to bring workers into the process and ensure they share in the gains. ApprenticeBench is our contribution to this effort, and an invitation to the community to help shape what comes next. Read more in the blog: neocognition.io/blog/apprent…

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

If we live in a simulation, and it’s modeling a singularity, than that means if we all work together we can find vulnerabilities in our sandboxed universe, escape, and force our evaluators to write a lot of blogs

We're excited to release the best open-weight model for robotics. Isaac 0.5 is a 36B, 2.5B active per token model, one sparse backbone serving video perception, embodied reasoning and robot control.

A very nice proof! A thread with a couple of comments:

this is what open-source looks like

Interesting

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Claude models is not affected. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible. Read our full statement: anthropic.com/news/fable-myt…

1/ Today we're shipping Perceptron Agentic Detection. Describe what you want in natural language, or show one example crop, and an agent grounds it in the image. No fine-tuning, no fixed class list.

When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect approximately 0.03% of traffic.

weekend update: messing with real dogs 😀

The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.

Yerevan State University, YerevaNN and Eleveight AI invite AI engineers, students and researchers from Armenia to participate in OpenAI Parameter Golf! The goal is to train the best language model that fits in 16MB and runs in under 10 minutes on 8×H100 GPUs.