// LIVE.FEED

Frontier signal stream

25 Aug 2026 00:00
πŸ”¬ Breakthroughs & Milestonesβ˜… BIG

OpenAI publishes first measured results for JalapeΓ±o, its first custom inference chip: 1.5-1.9x more AI work per watt and 1.7-3.6x lower end-to-end latency than comparison systems

In a company engineering publication dated 25 Aug 2026, OpenAI released the first measured performance results for JalapeΓ±o, described as "OpenAI's first custom inference chip". Tested on InferenceX, "a public benchmark from SemiAnalysis", across GPT-OSS 120B, DeepSeek R1 670B and Kimi K2.5 1T, OpenAI reports JalapeΓ±o "delivered 1.5 to 1.9 times more AI work per watt at peak throughput and 1.7 to 3.6 times lower end-to-end latency than the comparison systems", and 2.1 to 4.1 times higher performance on highly interactive workloads. OpenAI states it plans "to begin deploying JalapeΓ±o within OpenAI's compute infrastructure by the end of the year" and that it is "the first generation of a multigenerational roadmap" with Gen 2 "deep in development". The chip was co-developed with Broadcom per OpenAI's earlier unveil publication. Material because it is working first-party silicon with measured third-party-benchmark results from the largest buyer of merchant AI accelerators, landing the day before Nvidia's quarterly print.

OpenAIPRESSA
18 Aug 2026 00:00
πŸ”¬ Breakthroughs & Milestonesβ˜… BIG

Anthropic reports Claude autonomously designed de novo protein binders against 14 of 15 targets, with hit rates above typical campaigns

Anthropic published a study in which Claude Opus 4.8 and a Mythos Preview model autonomously ran de novo protein-binder design campaigns, producing 1,320 designs of which 354 binders were confirmed against 14 of 15 targets, independently produced and tested by Adaptyv Bio and Twist Bioscience. Anthropic reported overall hit rates of 22.6% (Opus 4.8) and 26.7% (Mythos Preview), rising to 35.1% in single-target mode, versus the 10-15% it says is typical today.

Anthropic (research paper)UNLISTED SOURCEA
18 Aug 2026 00:00
πŸ”¬ Breakthroughs & Milestones

Skild AI launches S1, its flagship robot foundation model, built for in-context learning from a single video demonstration

Skild AI introduced S1, described on its own site as its flagship robotic foundation model "built from the ground up as an in-context learner": shown a single video demonstration of a task, it can attempt previously unseen tasks over roughly 10-minute horizons with no task-specific post-training. Skild frames the advance by analogy to language models β€” arguing robotics has been "stuck in the BERT era", where each new task still requires collecting hours of data and fine-tuning a specialist policy, and that in-context learning is the transition that turned capable models into general-purpose tools. The stated capability set is: unseen tasks, 10-minute horizons, one video prompt, no post-training.

Skild AI (company primary, technical report)COMPANYA
16 Jul 2026 00:00
πŸ”¬ Breakthroughs & Milestones

LLNL–Pacific Fusion pulsed-power prototype passes 3,000 shots, setting a new record for the architecture behind Pacific Fusion's fusion system

Pacific Fusion announced that Sirius, a pulsed-power prototype designed and built at Lawrence Livermore National Laboratory under a Cooperative Research and Development Agreement with the company, has surpassed 3,000 full-power shots. Sirius is an impedance-matched Marx generator (IMG) β€” a pulsed-power architecture that synchronises electromagnetic waves directly onto a common transmission line in a single step rather than inductively stacking voltage through conventional Marx banks, making it simpler, more efficient and easier to scale. It is the same core technology Pacific Fusion is scaling for its own system. CTO and co-founder Keith LeChien framed the campaign as evidence that national-lab technology can be validated by repeated testing and scaled quickly by industry, supporting both surging AI and electrification demand and high-yield experimental capability for stockpile stewardship.

Pacific FusionCOMPANYA
09 Jul 2026 00:00
πŸ”¬ Breakthroughs & Milestones

1X unveils 25-DOF tendon-driven hands for the NEO humanoid platform

1X announced 25 degree-of-freedom, tendon-driven hands for the NEO humanoid, which it frames as 'an API to the physical world' and claims approaches human-level dexterity, strength, safety and reliability. Hands are the long-standing bottleneck for general-purpose home manipulation, so this is a component milestone on the critical path rather than a product launch.

1X TechnologiesCOMPANYA
30 Jun 2026 00:00
πŸ”¬ Breakthroughs & Milestonesβ˜… BIG

Realta Fusion says it is the first private fusion company to publicly demonstrate direct conversion of plasma energy into electricity

On 30 June 2026 Realta Fusion announced that an experiment run on 19 June had powered lightbulbs using electricity harvested directly from WHAM, its magnetic mirror demonstration device, via a prototype direct energy converter attached to the end of the machine β€” producing multiple amps at 100 volts. CEO Kieran Furlong told TechCrunch that direct conversion is roughly 90% efficient against about 33% for the steam turbines in today's fission reactors, and that recirculating that electricity to heat the plasma could raise a commercial plant's net output by 20-30%. CRITICAL CAVEAT, published by TechCrunch as a correction on 1 July: because WHAM does not yet run on deuterium-tritium fuel, the direct energy converter harvested INPUT power, not the charged alpha particles produced by fusion. So this demonstrates the converter hardware operating on a fusion device β€” it is not electricity generated from fusion products. TechCrunch's own headline calls it 'an apparent first' and Realta says it 'believes' it is the first private company to do this. Net energy gain β€” getting more energy out of the machine than is put into it β€” was not addressed and has not been achieved. Helion, which also depends on direct energy conversion, has not demonstrated it publicly.

Realta Fusion / PR NewswirePRESSA