An evaluation comparing GPT-6 Astra to GPT-5.6 Sol on software test/verification tasks found Astra achieved over 99% full test correctness on a Zstd implementation task at medium effort for about $1, versus Sol's 63-77% correctness at $3.50-$5.50. The Zstd eval has become saturated for Astra, nearly maxing out regardless of testing technique used, so evaluators also tried an IMAP RFC implementation task that had been too difficult for Sol to attempt meaningfully.
danluu.com
· 2026-11-16
SpaceX successfully caught the upper stage of its Starship rocket using the 'chopstick' arms of its launch tower after the vehicle completed a 22-hour orbit and deployed 26 Starlink satellites. The rocket's lower stage, however, was intentionally destroyed above the Gulf of Mexico after its descent was aborted.
nature.com
· 2026-11-16
MIT researchers showed that OpenAI's GPT-6 Astra model can interpret the internal numerical values of other neural networks well enough to predict what different components do, and can then edit those networks to improve performance, including on math tasks. The team, led by Rikiya Takehi, described the findings in a preprint posted on November 2.
nature.com
· 2026-11-13
A writer used OpenAI's GPT-6 Sol model to teach a virtual simulation of Pollen Robotics' Microduck robot to walk around and identify a bengal cat, with plans to make it dispense treats via a 3D-printed head attachment once a physical unit ships. The process reportedly took only a few hours, compared to what the author says would have previously taken weeks. Separately, UC Berkeley professor Ken Goldberg and students have explored using OpenAI's Astra model, released last September, for generating the detailed 3D simulations needed to train robots.
wired.com
· 2026-11-11
OpenAI said it will shut down the Advanced and Standard voice modes in ChatGPT on December 11, 2026, consolidating all voice interactions into its newer GPT-Live model. The company said voice usage has nearly doubled since GPT-Live launched earlier this year, with most users already on that mode. Live video and screen sharing will remain on Advanced voice until GPT-Live supports those features.
androidauthority.com
· 2026-11-11
A researcher analyzing MIT CSAIL's 'Sparks of Weight-Space Programming' study found that GPT-6 Astra's 8-hour edit to Qwen3.5-0.8B's weights, which lifted MATH-500 accuracy from 7.9% to 48.1%, was effectively solving a linear program through trial and error. The author reports that running the same optimization directly as a linear program takes about a minute per solve and changes 8 times fewer weights than Astra's search. The underlying fix addressed repetitive 'thinking loops' in the small model, also boosting IFEval scores from 21% to 62%.
nathan.rs
· 2026-11-11
Muse has released Ember, described as its smallest and most efficient model, built for tool calls, coding subtasks, routing, extraction and long document parsing. According to benchmarking data cited by the company, Ember matches Claude Haiku 5.5's top score at roughly a third of the cost and outperforms GPT-6 Luna at every reasoning setting tested.
research.meta.ai
· 2026-11-09
New testing data compares how different hardware handles local large language models, measuring output in tokens per second. Nvidia's RTX 5090 generated 205 tokens per second on gpt-oss-20b, Apple's M4 chip produced about 24 tokens per second on a 7B model, and Nvidia's DGX Spark ranged from 38 tokens per second on Llama 3.1 8B down to 4.42 on its 70B version. Separate hands-on testing on a 2024 Asus Zenbook Duo showed Intel's OpenVINO toolkit offloading work to an NPU kept the laptop responsive, versus heavy CPU and memory strain when running the same model through Ollama.
engadget.com
· 2026-11-09
An innovation professor's essay draws on the 2017 'Attention Is All You Need' paper, which introduced the Transformer architecture behind GPT and the generative-AI boom, to argue that enterprise AI has so far only changed individual work. The piece contends a comparable architectural breakthrough is still needed before AI can transform entire organizations rather than just personal productivity.
fastcompany.com
· 2026-11-08
OpenAI has made outbound SIP phone calling generally available for GPT-Live, letting any organization's backend dial phone numbers through a SIP trunking provider without requiring an incoming-call webhook. Developers configure a SIP trunk pointing to OpenAI's endpoint and create a session to place calls, with TLS for signaling and SRTP mandated for call audio.
developers.openai.com
· 2026-11-06
OpenAI launched GPT-6 Cyber, a variant of its newest flagship model built specifically for cybersecurity defenders, giving early access to security teams at companies including Mozilla and Palo Alto Networks. The rollout came the same week Anthropic also pushed updates to Claude, as both companies continue adding incremental capabilities to their chatbots rather than full public releases of new frontier models.
cnet.com
· 2026-11-06
OpenAI announced GPT-6 Cyber, an AI model built for vulnerability research, which uncovered 81 unknown high-severity Firefox flaws and helped patch many of them. In Linux kernel testing, the model flagged 2,308 potential issues, reproduced 1,724 of them, and built privilege-escalation exploits for 948 distinct bugs. OpenAI rated the model's cybersecurity capability as 'Critical,' up from 'High' for its predecessor, and restricted access to select customers in its Daybreak Red program.
bleepingcomputer.com
· 2026-11-06