Samwise Tech/AI/Robotics Newsletter
Wednesday, June 11, 2026
Neura Robotics raises $1.4B Series C from Nvidia, Amazon and Tether, reaching $7B valuation
German humanoid robotics startup Neura Robotics closed a $1.4 billion Series C round—the largest single raise in humanoid robotics history—anchored by Tether Holdings, with Nvidia, Amazon, Qualcomm, Robert Bosch, and Schaeffler also participating. The Metzingen-based company, valued at approximately $7 billion, builds the 4NE-1 humanoid robot targeting manufacturing and logistics environments, along with a factory transport robot and a domestic robotic arm. The company said its existing orders and strategic deployment pipeline already exceed $1 billion. Proceeds will fund robot manufacturing scale-up, its AI software platform, and global distribution infrastructure.
Sources: Robotics & Automation News
GPT-5.5 tops new Agents’ Last Exam leaderboard with 24% pass rate, edging Anthropic’s Claude Fable 5
UC Berkeley’s Center for Responsible, Decentralized Intelligence launched Agents’ Last Exam (ALE), a rigorous benchmark testing whether AI can execute economically valuable, long-horizon professional workflows across 55 non-physical industry sub-domains from the U.S. federal occupational taxonomy. On the inaugural leaderboard, OpenAI’s GPT-5.5—operating through the Codex harness—scored 24.0%, edging Anthropic’s Claude Fable 5 at 22.0%. The benchmark was developed with an advisory committee of over 300 domain experts and is designed to measure real-world autonomous task completion rather than academic test performance. Researchers and developers can submit models directly for public evaluation on the ALE platform.
Sources: VentureBeat
Decart launches Oasis 3, an interactive world model for photorealistic driving simulation, via API at $0.02 per second
Decart, a two-year-old AI startup valued at approximately $4 billion following a $300 million raise, unveiled Oasis 3, a real-time interactive world model that generates photorealistic, multi-camera driving environments for autonomous vehicle testing and robotics training. The model is available via API at $0.02 per second, with enterprise pricing on request. Oasis 3 is built on Decart’s DOS (Decart Optimization Stack), which the company says makes its inference costs more than an order of magnitude cheaper than competitors. Unlike prior world models, Oasis 3 supports infinite scenario generation, though scene consistency degrades over extended sessions and physics simulation remains an open research problem.
Sources: TechCrunch
Waymo publishes Reference Driver model in Nature Communications to benchmark robotaxi safety against human behavior
Waymo, in collaboration with TU Delft, published research in Nature Communications introducing the Reference Driver, a new computer model that more accurately benchmarks autonomous vehicle safety against human driving behavior. Unlike prior industry models that focused only on last-second reactions, the Reference Driver uses an active inference framework to simulate the internal “surprise” a human driver experiences during a conflict scenario, enabling pre-crash behavioral modeling. The company is making the research code available under an academic, non-commercial license. The work comes as Waymo faces ongoing federal investigations into a January 2026 incident in Santa Monica in which one of its robotaxis struck a child.
Sources: TechCrunch
Researchers train HRM-Text foundation model from scratch for $1,500, challenging conventional AI training economics
Researchers published a paper describing how they trained HRM-Text, a foundation model, from scratch in just 1.9 days on a 16-GPU cluster at a total estimated compute cost of roughly $1,500. Pretraining foundation models from scratch typically costs millions of dollars and is reserved for well-funded labs, making this result noteworthy. The team reports that HRM-Text achieved competitive scores using 100 to 900 times fewer training tokens and 96 to 432 times less estimated compute than comparable models including Qwen, Gemma, and Llama. The work highlights how efficiency improvements and better training regimes continue to reduce the cost barrier for AI research.
Sources: VentureBeat
SpaceX alumni launch Ambrosia Energy to deliver solar-plus-battery power plants at $100/MWh for AI data centers
Sara Spangelo and Ben Longmier, co-founders of satellite startup Swarm (acquired by SpaceX in 2021), emerged from stealth with Ambrosia Energy, a startup building solar-plus-battery power plants designed to undercut natural gas on both cost and construction time. Targeting AI hyperscalers, Ambrosia promises $100 per megawatt-hour combined solar and storage, with plants deployable in under 12 months from contract to grid-on—versus the five-to-seven-year backlog facing gas turbine orders. The company began constructing its first site in West Texas in January 2026, with sections already operating at 100% capacity. DFJ Growth has invested; the investment size was not disclosed.
Sources: TechCrunch
Jedify raises $24M Series A to build enterprise context graphs that help AI agents understand business data
New York-based startup Jedify raised $24 million in a Series A round led by Norwest, with participation from S Capital VC, Cerca Partners, Oceans Ventures, and strategic investor Snowflake Ventures. Jedify’s platform connects to enterprise data sources—including databases, data warehouses, SaaS tools, Slack channels, and meeting recordings—to build a real-time “context graph” that gives AI agents the business-specific knowledge they need to perform autonomously. Unlike semantic layers or metadata catalogs, the company’s multi-dimensional context graph captures relationships across entities, data, people, permissions, and workflows, while inheriting row-, column-, and table-level access rules. Total funding reaches approximately $33 million.
Sources: TechCrunch
Tech Pulse
Top Frontier Models: Claude Mythos 5 (95.5%) | Claude Fable 5 (95.0%) | Claude Opus 4.8 (88.6%)
Top Open Source Models: MiniMax M2.5 (80.2%) | GLM-5 (77.8%) | DeepSeek V4-Pro (76.4%)
Top Small Models (15–50B): Qwen3-32B (63.4%) | Gemma 3 27B (56.8%) | Phi-4-Reasoning 14B (54.2%)
Top Edge Models (0–15B): Qwen3 8B (46.5%) | Phi-4 Mini (43.1%) | Gemma 3 4B (39.7%)
AI Leaders: Nvidia $5.0T | Alphabet $4.6T | Microsoft $3.1T
Robotics Leaders: Intuitive Surgical $175B | ABB $165B | Fanuc $32B
Curated by JD · samwise.agency