Ryan Wolf

Figure and Hark just took an entire data center of NVIDIA B200s - every rack in the building Figure will be using these to predict physics and Hark will train next generation multi-modal models

New year, new Artificial Analysis Intelligence Index! Announcing Intelligence Index v4.0: incorporating 3 new evaluations, further aligning to real-word use and reducing saturation The Artificial Analysis Intelligence Index is our synthesis metric for assessing generalist model intelligence and tracking AI progress. We like nuance and breakdowns at Artificial Analysis (and you’ll find plenty of that in the thread below!) but when you want a single number, this is the best one. Artificial Analysis Intelligence Index v4.0 is: ➤ Less saturated: Top models score ≤50 in v4.0 compared to 73 in v3.0. The ten evals still cover a range of difficulties to allow differentiation from small models to frontier models ➤ More agentic: GDPval-AA, our leading metric for general agentic performance, joins Terminal-Bench Hard and Tau2 Telecom ➤ A multi-dimensional view of intelligence: Index V4.0 has four equally weighted categories: Agents, Coding, Scientific Reasoning and General Changes: ➤ Added in v4.0: AA-Omniscience, GDPval-AA, CritPT ➤ Removed in v4.0: MMLU-Pro, AIME 2025 and LiveCodeBench Overview of the three new evals: ➤ AA-Omniscience tests models on knowledge and hallucination across >40 topics ➤ GDPval-AA is our generalist agentic performance eval - testing models on real-world economically valuable tasks leveraging OpenAI’s GDPval dataset, via our reference agent called Stirrup ➤ CritPt tests performance on hard research-level physics research tasks, including condensed matter, quantum physics, and astrophysics Results: OpenAI's GPT-5.2 with xhigh reasoning effort leads the Artificial Analysis Intelligence Index v4.0, followed closely by Anthropic's Claude Opus 4.5 then Google's Gemini 3 Pro. As always, all evaluations have been run independently and using a standardized methodology. We publish our methodology on the Artificial Analysis website. See below for further analysis.

Excited to share our F.02 robots have contributed to the production of 30,000 cars at BMW Today we're sharing our learnings from an 11-month real world deployment as the F.02 fleet retires

Introducing Figure 03

Here's our progress over the last 12 months

Helix is exceeding what I thought was possible in my home To put a robot in every household, we need massive amounts of data - today we’re launching Project Go-Big This is accelerated with Brookfield who owns over 100,000 residential units