AI Daily · Nov 7, 2026 · Sat
ICREAT AI NEWS
Thinking Machines releases Inkling 2 Small, weights pending release
Thinking Machines releases Inkling 2 Small, scoring 31 on the Artificial Analysis Intelligence Index v5.0.1, 18 points higher than Inkling Small and on par with Gemini 3.8 Flash (high).
Product releases/updates
OpenRouter AnnouncementsFirst partyOpenRouter releases a new generation of Auto and Auto Code routers
OpenRouter releases a new-generation Auto Router and adds Auto Code (openrouter/auto-code) for coding scenarios; the flagship general-purpose version remains openrouter/auto.
Cursor ChangelogFirst partyCursor launches real-time agent voice calls
Cursor introduces real-time agent voice calls: users can describe changes verbally, ask what the agent is doing, and hear progress updates as tasks advance. The feature covers the Agents Window, IDE, web, and Cursor for iOS, is powered by Grok Voice, offers multiple voices such as Ara, Eve, Leo, Rex, and Sal, and retains zero data when Privacy Mode is enabled. Voice calls and dictation are independent; dictation still handles converting speech into text in the chat input box.
Amp ChronicleFirst partyAmp launches spaces, where a single agent derives and coordinates multiple agents
Amp releases spaces, a new way to collaborate with agents and teams: a space is a conversation with a single agent that creates and coordinates multiple agent threads for you, so you don't have to organize threads yourself. The Amp team used a space to collaboratively develop Spaces v1, ship and troubleshoot service accounts, and conduct a security review that split the codebase into 16 areas.
Industry news
The DecoderAnthropic's $1.5 trillion IPO hits a snag as investors question the valuation
Anthropic's IPO has stalled after investors objected to the $1.5 trillion valuation, with some potential backers doubting whether the roadshow originally scheduled to start next week can proceed as planned. The company had targeted a valuation of up to $2 trillion in September; now some investors consider even $1.5 trillion too high, and some have even floated just $1 trillion.
Democratic midterm victory sparks AI investment fears, global stocks plunge
After a landslide Democratic victory in the U.S. midterm elections, fears that the AI investment boom will falter sent global stocks sharply lower. More than $3.2 trillion was wiped off U.S. stock market value, the Nasdaq fell 5.5%, Nvidia dropped over 9%, SanDisk fell 22%, and Micron and AMD each fell 17%. Asia-Pacific markets fell in tandem, with the Nikkei down nearly 6%, South Korea's KOSPI down 6.7%, and Taiwan's TAIEX down 4.5%.
The DecoderPlus 1 sources reportingMETR audit finds Anthropic may have crossed its own AI acceleration warning threshold in July
METR's internal audit of Anthropic shows that its model test scores have been improving about 1.9 times as fast since October 2025, covering two years of past progress in one year on the composite AECI metric, and Anthropic internally believes the warning threshold may have been reached on July 15.
Paper research
vLLM BlogFirst partyREINFORCE Zero: after aligning kernels, asynchronous RL trains stably at lag 64
The paper REINFORCE Zero from NVIDIA Research and Inferact proposes first aligning the kernels on the rollout and training sides, then using uncropped token-level importance sampling to correct for weight lag, achieving 2.40 times the training-step throughput of synchronous training (68.55 vs. 164.77 hours) with 16 GPUs and 12K updates.
Tips and opinions
The cost of AI breaking into enterprise networks falls 99% in five months, shifting security budgets toward a compute race
Tomasz Tunguz notes that the cost of AI breaking into enterprise networks fell 99% in five months: BIS estimated that in April, Claude Mythos Preview and GPT-5.5 needed $5k to $10k to autonomously breach a simulated UK enterprise network, while by August Zhipu's open-weight model GLM-5.3 reached a similar attack success rate for $90.
Google Developers BlogFirst partyGoogle research: how legacy TODOs in codebases distract coding agents
Based on distributed tracing across thousands of engineers and hundreds of millions of agent steps, Google quantified how TODO(bug/...) comments in codebases interfere with coding agents: 13% of sessions exposed agents to such references, 6% of those triggered follow-up actions, and 83% of that 6% were not requested by developers.