Over 300,000 designs in 64 hours; 17 of the 211 proteins sent for lab testing met the binder criteria.
Anthropic says Claude has made “an early step toward finding new treatments” for scorpion stings. Working with @citrini, Anthropic asked Claude to design proteins that bind to five scorpion toxins, covering the part of the toxin that latches onto cells. Over the next 64 hours, Claude generated over 300,000 designs. Anthropic sent 211 to @adaptyvbio and @TwistBioscience for lab testing, and 17 met the criteria for a binder.
More than 40x the size of Mercury 1, over 1,000 tokens a second and a 1M-token context window.
Inception has launched Mercury 3, which it calls “the most capable diffusion LLM released to date.” The company says it outperforms GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks while running at over 1,000 tokens per second. Mercury 3 is more than 40x the size of Mercury 1 and, by Inception’s count, 3x more intelligent across standard benchmarks.
A friend at a traditional company in Austria: “everything takes me only 5 seconds now.”
@levelsio says a friend with a regular software job at a traditional company in a small Austrian town got “fully AI-pilled” today. “It’s crazy, everything takes me only 5 seconds now,” the friend wrote in their group chat.
A user-facing coordinator and independent agents passing messages; he wants multiplayer AI past the group chat.
Ethan Mollick calls the new Claude Projects a nice interface for multi-agent work: a user-facing coordinator model and independent agents running threads that pass messages around. It “takes some fiddling but quite powerful,” he says, and has no relation to the old Projects.
@sudoingX runs his Rust harness with offrail 27b on a laptop; it rewrote its own image skill.
@sudoingX says he is “falling for” his own agent harness. xpress agent is written in Rust, ships as one binary under 10 mb, uses 14 mb of RAM for the whole loop and finds his local model by itself. offrail 27b is the model behind it, running on his laptop. It writes its own skills as it works, and on one run it rewrote its image skill without being asked.
In @cybersec’s VulnHunter eval, DS4.1-1017 turned up one new vulnerability in a production Go repo.
@MiaAI_lab shares results from @cybersec: an abliterated DeepSeek V4.1 Flash 1017 found one new vulnerability that a frontier model did not discover. @cybersec runs an operational pass against a pinned Go repo for a production app with known vulnerabilities, with Grok as the frontier baseline for all open-weight evals.
A 30B open-weight model for agentic tasks on a single consumer GPU at 4-bit precision.
@kimmonismus flags Meta’s technical report on Muse Glimmer, a 30B open-weight model built for agentic tasks on consumer hardware. Meta says it runs at practical speeds on a Mac or PC with a single consumer GPU at 4-bit precision.
@swyx surfaces DeepSeek’s Zhihu figures, including $38 per card per hour in revenue at off-peak prices.
@swyx says DeepSeek posted “way more alpha” on its Zhihu page than on X. He quotes DeepSeek: the DeepInfer-based inference service for DeepSeek V4.1 Flash has a single-cluster throughput of 155 trillion tokens a day, and overall revenue is $38 per card per hour, based on off-peak prices.
Open weights under the MIT license; @MiaAI_lab posted an EXL3-EAQ v1 quant for one DGX Spark or RTX PRO 6000.
Ornith released Ornith-1.5-122B, open weights under the MIT license, and says it runs on a single NVIDIA DGX card. @MiaAI_lab congratulated the team and posted a day-zero EXL3-EAQ v1 quant for one DGX Spark or RTX PRO 6000.
1,151 Elo on Artificial Analysis’s AA-Image-Editing v2.1, one point behind Google’s Nano Banana 2.1.
@kimmonismus says Krea’s published results put an independent lab right alongside Google and Microsoft in image editing. Krea 3 Ultra scores 1,151 Elo on Artificial Analysis’s AA-Image-Editing v2.1.
The stack behind DeepSeek V4.1 Flash is public; for now it runs on NVIDIA SM100 and Huawei Ascend 950.
DeepSeek open-sourced DeepInfer, the inference engine behind its own production services. @MiaAI_lab lists draft tokens for faster output, custom kernels, skipping repeated work on long prompts, and the same output for the same input every time.
@sudoingX’s 2.70-bit v1.2 hits 141.3 tok/s on structured output; @MiaAI_lab says EAQ v1 is 11% to 38% closer.
@sudoingX rebuilt his DeepSeek V4.1 Flash weights on the 1017 update. His v1.2 runs on 2x DGX Sparks at 141.3 tok/s on structured output, 100.7 on code and 59.2 on prose.
Video Rebirth’s open-source add-on runs about 3x faster; a 175-second clip now takes about 60.
Video Rebirth open-sourced HyperFlow, a speed-up for the MiniMax H3 video model, @kimmonismus reports. The add-on keeps the same workflows and replaces the 49-step schedule with a trained 8-step one, running about 3x faster with sound generated alongside the video.
A joint certification would combine SOC 2 and AIUC-1 examinations into a single independent review.
The @AICPA and the Artificial Intelligence Underwriting Company announced a plan to launch a joint assurance certification, Accounting Today reports. It combines the two organizations’ SOC 2 and AIUC-1 examination processes into a single comprehensive independent review.
A CAQ survey finds adoption ahead of governance; Ascend’s David Wurtzbacher says firm structure may need rethinking.
Accounting Today says AI adoption “is in the driver’s seat” while governance “is still riding in the back seat,” and looks at what the CAQ’s latest survey means for the rest of 2026. On an Accounting Today podcast, Ascend founder and CEO David Wurtzbacher suggests that getting the full capabilities of artificial intelligence may require rethinking the entire structure of the accounting firm.
At $1.4T+, per @kimmonismus, with $50B in annualized revenue and $70B targeted by year-end.
OpenAI is reportedly rushing a ~$30B raise at $1.4T+ so it closes before Anthropic’s IPO in mid-November, @kimmonismus reports. He puts OpenAI at $50B annualized revenue now, with $70B targeted by year-end.
OpenAI’s former research chief introduces a startup now valued at $1.5B.
Bob McGrew, OpenAI’s former research chief, just introduced Arda, an AI manufacturing startup now valued at $1.5B after a $230M round, The Rundown AI reports, via CNBC. McGrew says Arda does for factory engineers “what Claude or Cursor does for programmers.”
DHS tells civilian agencies to remove DeepSeek and products built on it; DeepSeek reportedly helps design Huawei chips.
The US Department of Homeland Security is ordering civilian federal agencies to remove DeepSeek and third-party products built on it, and to stop buying them, @kimmonismus reports. He reads it as a response to China’s proposed export controls on certain AI technologies, part of a broader push toward AI sovereignty.
A watchdog, a CERN-style body for the labs, or one company that can “melt everyone else’s GPUs.”
Sam Altman described three paths to OpenAI staff, according to The Information via @kimmonismus. One is an international watchdog modeled on the atomic energy agency.
Anthropic’s co-founder points to limits on “the ceiling of intelligence” and “a small number of years.”
Anthropic co-founder Jack Clark went on record at the IMF about putting a ceiling on AI intelligence, The Rundown AI reports. Moderator Raghuram Rajan asked when humans hand AI the reins.
New Microsoft Sovereign Cloud capabilities; Saint-Gobain uses Microsoft Discovery on construction materials.
Satya Nadella says Microsoft announced new capabilities across Microsoft Sovereign Cloud in Paris, expanding choice and control for organizations running their most sensitive workloads across public and private cloud. He also met the Saint-Gobain team, which is using Microsoft Discovery to help accelerate the development of new, more efficient construction materials.