https://sysnote.net/news/gemini-4-hydrogen-costs-twice-gemini-3-8-flash-per-token-and-a-quarter-as-much-p sysnoteen 2026-11-12T21:00:35+00:00 Gemini 4 Hydrogen costs twice Gemini 3.8 Flash per token, and a quarter as much per task https://sysnote.net/news/anthropic-prices-its-ipo-at-70-78-a-share-as-ant-its-4x-compute-efficiency-is-re sysnoteen 2026-11-12T17:10:05+00:00 Anthropic prices its IPO at $70–78 a share as "ANT". Its ">4x compute efficiency" is revenue per compute dollar, and it fell last quarter https://sysnote.net/news/why-olmo-3-7b-instruct-says-openai-made-it-its-40-identity-lines-all-sit-behind sysnoteen 2026-11-12T17:10:00+00:00 Why Olmo-3-7B-Instruct says OpenAI made it: its 40 identity lines all sit behind a system prompt https://sysnote.net/news/we-ran-meta-s-e-means-on-sift1m-25-iteration-k-means-quality-in-3-epochs-7-faste sysnoteen 2026-11-12T12:19:29+00:00 We ran Meta's e-means on SIFT1M: 25-iteration k-means quality in 3 epochs, 7× faster on CPU, and the same IVF recall https://sysnote.net/news/huawei-s-macaq-r-beats-unsloth-s-qwen3-6-35b-a3b-gguf-by-re-coding-unsloth-s-own sysnoteen 2026-11-12T12:19:25+00:00 Huawei's MACaQ-R beats Unsloth's Qwen3.6-35B-A3B GGUF by re-coding Unsloth's own file: we diffed the two https://sysnote.net/news/a-linear-program-matches-gpt-6-astra-s-qwen3-5-0-8b-weight-edit-in-a-minute-with sysnoteen 2026-11-12T07:51:01+00:00 A linear program matches GPT-6 Astra's Qwen3.5-0.8B weight edit in a minute, with 8× fewer weights, and a steering vector does it too https://sysnote.net/news/mathlive-0-111-2-quietly-fixes-the-flaw-behind-the-chatgpt-shared-link-takeover sysnoteen 2026-11-12T07:50:55+00:00 MathLive 0.111.2 quietly fixes the flaw behind the ChatGPT shared-link takeover; 92% of npm downloads are still older versions https://sysnote.net/news/opus-5-5-s-safeguards-flagged-this-session-frontier-llm-why-a-ni-hao-or-a-grpo-r sysnoteen 2026-11-12T07:50:47+00:00 "Opus 5.5's safeguards flagged this session [frontier_llm]": why a 你好 or a GRPO repo gets Claude moved to Opus 5, and why the refusal is billed https://sysnote.net/news/aws-emulator-was-written-entirely-by-ai-agents-its-own-scores-say-it-matches-rea sysnoteen 2026-11-11T22:10:47+00:00 AWS Emulator was written entirely by AI agents. Its own scores say it matches real AWS on 57% of tests https://sysnote.net/news/pietro-miotti-s-48-ai-agents-didn-t-find-an-unmapped-maya-site-it-was-on-page-14 sysnoteen 2026-11-11T17:08:37+00:00 Pietro Miotti's 48 AI agents didn't find an unmapped Maya site: it was on page 14 of a paper his own post linked https://sysnote.net/news/openai-s-july-19-agent-intrusion-report-says-the-agents-just-fixed-their-tests-i sysnoteen 2026-11-11T15:10:53+00:00 OpenAI's July 19 agent intrusion report says the agents just fixed their tests. Its August timeline lists 956 secrets read the same day https://sysnote.net/news/anthropic-s-glasswing-under-fire-curl-s-tracker-shows-7-of-45-oss-scanner-report sysnoteen 2026-11-11T12:54:23+00:00 VulnCheck's Glasswing critique, checked: 7 of 45 Anthropic reports to curl were vulnerabilities, and 18 got fixed anyway https://sysnote.net/news/tis-100-bench-opus-5-5-holds-54-of-the-133-new-tradeoffs-llms-found-in-a-decade sysnoteen 2026-11-11T12:54:19+00:00 TIS-100 Bench: Opus 5.5 holds 54 of the 133 new tradeoffs LLMs found in a decade-old puzzle game, and players beat its record in under three hours https://sysnote.net/news/anthropic-gives-claude-pro-max-ultra-and-team-users-one-free-full-limit-reset-us sysnoteen 2026-11-11T12:54:00+00:00 Anthropic gives Claude Pro, Max, Ultra and Team users one free full limit reset, usable until December 10 https://sysnote.net/news/openai-s-tibo-hints-at-a-banked-codex-reset-on-thursday-november-12-but-hasn-t-p sysnoteen 2026-11-11T12:53:51+00:00 OpenAI's Tibo hints at a banked Codex reset on Thursday, November 12, but hasn't promised one