Morning Edition · Monday, October 5, 2026
Tech
Anthropic and OpenAI price their near-flagship models at $2 and $10 per million tokens while open-weight Qwen runs on a gaming GPU
Claude Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, ahead of Opus 5.5's 66.4%, at an unchanged list price. Most of the cost reduction comes from fewer tokens and cheaper cache reads.
Tech
Hinton says Trump is "badly advised" on AI after the White House replaced rules with a voluntary self-policing accord
Executives signed the accord at a September 29 luncheon. It commits their companies to internal controls, an outside assessor and a board committee. No federal agency enforces it.
Tech
New papers show jailbreaks and GraphRAG poisoning work by composing benign parts, a pattern OpenAI says it saw in a reasoning-extraction campaign
OpenAI says accounts linked to China's Moonshot AI decoded its hidden reasoning by moving encrypted reasoning traces into new conversations. More than 4,000 accounts sent 16,000 requests over two days.
Tech
Sonnet 5.5's jump from 10.3% to 70.6% on Terminal-Bench 4.0 tests how far vendor agent evals can be trusted
Two new arXiv benchmarks argue for scoring agents in closed loops and on calibrated probabilities. XiangqiBench has agents play full games against a responding opponent. The Turkish-language HakemBench scores a probability for every option.
Tech
MIRROR paper defends multi-agent systems against agent-in-the-middle tampering as harnesses hand routing and injection checks to fast single-pass models
A companion arXiv study audits the single-pass "System-1" models that harnesses use to choose tools and flag injected instructions. OpenAI announced a Decisions API for this same layer of bounded decisions at its DevDay event.
Tech
Two papers find supervised fine-tuning teaches surface style and can widen a falsification gap that reinforcement learning closes
One study explains why supervised fine-tuning (SFT) causes more forgetting than reinforcement fine-tuning (RFT), even when every demonstration is correct. The other shows that imitation makes models worse at disproving false theorems.
Macro
Anthropic's 2030 scenarios put US output up to 32.4% above a no-AI path while labor's share of income falls to 45.2%
The typical answers in Anthropic's survey of 10,980 Americans imply an economy about 10% larger than it would be without AI, with unemployment near 5%. Those answers sit close to the middle of Anthropic's three scenarios.
Tech
Meta sells its closed-weight Muse Spark 1.3 at $1.25 and $4.25 per million tokens through an OpenAI-compatible API
Meta prices its fourth Muse Spark release in five months below the new $2 and $10 tier from Anthropic and OpenAI. The model's weights stay closed, and it offers a 1-million-token context.
Tech
Department of Energy labs run Meta's SAM 3 and DINOv3 on 300 A100s to cut beamline image segmentation from a month to about 15 minutes
The SYNAPS-I project, led by Lawrence Berkeley National Laboratory under the Department of Energy's Genesis Mission, fine-tunes Meta's open vision models for detectors that now capture 100,000 images per second.
Tech
Runway's Project Continuum generates operating-system interface regions frame by frame from a video world model
The research prototype has no release date. It builds on Solaris, an interface world model that Runway introduced on August 31 on top of its Gen-4.5 video model.
Tech
World Health Organization staff cut the daily Ebola situation report in Congo from a full day to under an hour with a Claude workflow
Anthropic says partners of the Coalition for Epidemic Preparedness Innovations (CEPI) also use Claude to run several outbreak models, where they once had time for one. Human experts make every response decision.
Tech
Musk says SpaceX's AI unit will rename itself SpaceXSI to match the White House's "super intelligence" order
The unit was formed in July after SpaceX's reported $1.25 trillion acquisition of xAI. The change would give it its second new name in about three months. It would also align a federal contractor's brand with executive-branch vocabulary.