Morning Edition · Monday, October 26, 2026
Tech
Recursive releases Flint, a free coding model with 2.5 billion active parameters that it says its own AI research system mostly built
The 39.8-billion-parameter mixture-of-experts model runs at about 131 tokens per second on an M4 Pro MacBook Pro, but every benchmark score Recursive has published comes from its own test runs.
Tech
Chinese AI firms offer up to 1.09 million yuan for deployment engineers as export-barred Nvidia B300 chips appear on China Telecom-linked systems
Postings for forward-deployed engineers in China rose 1,523% year on year, while a scan of unsecured GPU monitoring endpoints found 32 Blackwell Ultra chips that United States rules bar from sale to China.
Tech
Anthropic says a Claude research model designed 318,484 protein candidates in 64 hours, and 17 passed laboratory binding tests
The model used 12 protein design tools against five scorpion neurotoxins, and Anthropic states plainly that it has not shown the binders neutralize venom in cells.
Tech
Anthropic prices Claude Opus 5.5 at 20% below Opus 5 per token and claims 66.4% on Terminal-Bench 4.0
Anthropic says the model costs 40% less to run on typical workloads, with cache reads down 60%, while every launch benchmark comes from its own evaluation table.
Tech
Berkeley Lab cut beamline image analysis from months to about 15 minutes using Meta's open-weight vision models
The five-laboratory SYNAPS-I pipeline runs Segment Anything Model 3 and DINOv3 on 300 Nvidia A100 GPUs, and the same two models sit inside a University of Pittsburgh assistive robotics platform funded with up to $41.5 million.
Tech
OpenAI's Decisions API bills input tokens only and returns a typed answer in about 150 milliseconds
The endpoint charges $0.10 per million input tokens with no output or cache charges, and OpenAI says it is roughly ten times faster than routing the same decision through the Responses API.
Tech
A mathematicians' association has 1,008 members pledged not to work with commercial AI companies
Of those, 654 have joined an AI-free caucus committing not to use AI models for research at all, including for finding proofs or posing questions.
Markets
Apollo's chief economist says AI has produced no financial payoff for United States companies outside technology
Revenue per worker is flat across the S&P 500 excluding its seven largest technology firms, whose revenue per employee has risen to roughly $270,000.
Tech
Anthropic commits 5% of its shares to a wholly owned fund, rising by 2% for every 50% gain in company value
Combined with pledges from cofounders and employees, Anthropic says about 20% of its equity is directed to public purposes, and none of the fund's shares may be used for lobbying or political giving.
Tech
Meta's 816-billion-parameter Muse Spark Open trails Chinese models on coding while leading them on agentic benchmarks
On Terminal-Bench 2.1 the model scores 76.4 against Kimi K3's 85.0 and GLM 5.3's 83.9, but Meta's table shows it ahead of both on tool-use evaluations including JobBench and MCP-Atlas.
Tech
A Lancet study found Google's AMIE matched physicians' final diagnoses 90% of the time across 98 patient consultations
Every conversation was monitored live by a board-certified internal medicine physician, and none met the predefined criteria for interruption.
Tech
Two papers argue that benchmark scores on interactive LLM tasks measure the evaluation design more than the model
One shows that overall accuracy can invert conclusions when comparing responses to different follow-up prompts, and the other finds that scored success in AI math tutoring shifts with the grading criteria chosen.