Apple has introduced LensVLM, built on Qwen3.5-9B-Base, which lets vision-language models scan compressed text-as-image inputs and selectively expand only relevant portions using learned tools rather than reading everything at full resolution. The model reportedly matches full-text accuracy at 4.3x compression and beats retrieval and compression baselines up to 10.1x compression across seven QA benchmarks, plus document and code tasks.
huggingface.co
· 2026-09-23
A developer testing an AI model at maximum effort settings found that most requests produced little or no chain-of-thought reasoning tokens. Even when extended reasoning occurred, it fell well short of the levels seen in the model's published benchmark results.
twitter.com
· 2026-09-21
Security firm Socket found that the Chrome and Firefox extension 'Twitch Enhanced Viewer | JeetBot,' which promises ad-blocking and forced 1080p playback, secretly captures users' Twitch OAuth session tokens. The credentials are routed through proxy servers run by JeetBot, a Russian-language streaming and chatbot service, where they land in plaintext request logs accessible to the vendor. The extension exempts ten hardcoded Russian-language channels from this behavior, and its own store listing admits earlier versions sent tokens to its servers outright.
bleepingcomputer.com
· 2026-09-14
Nasdaq's venture arm is putting $100 million into Payward, the parent company of crypto exchange Kraken, valuing it at $21 billion. The two firms plan to build infrastructure for issuing and trading tokenized equities, blockchain-based digital representations of publicly traded stocks, with a target launch in the second quarter of 2027. Payward will also adopt Nasdaq's market-surveillance technology across its trading platforms as it expands beyond crypto into stocks and derivatives.
cnbc.com
· 2026-09-10
Meta reportedly built an internal leaderboard called Claudeonomics that ranked employees by AI token consumption, awarding titles like 'Token Legend' to top users who burned through over 60 trillion tokens in a month. Similar tracking has surfaced at Disney, JPMorgan, KPMG, Amazon, and Accenture, with some firms incorporating AI usage into performance reviews.
fastcompany.com
· 2026-09-09
Nitter, the open-source alternative front-end for Twitter/X, now has more publicly working instances than it did before a series of takedowns hit the project. New volunteer-run instances have come online, alongside guidance for operators on securing session tokens and handling DMCA and other legal complaints.
codeberg.org
· 2026-09-05
A critical authentication bypass, CVE-2026-82329, in self-managed JFrog Artifactory deployments is being actively exploited to mint fraudulent administrator tokens without any prior authentication. The flaw sits in Artifactory's default configuration, and watchTowr researchers say attackers are already using it in the wild. JFrog patched the issue on August 28 across several version branches, and confirmed its cloud-hosted environments were never at risk.
bleepingcomputer.com
· 2026-09-02
After burning through a Claude usage quota in just 10 minutes, a developer built 'tare,' a skill that lets Claude Code analyze its own local request logs to explain where tokens are going. Users install it with a single command and then ask plain-English questions about spikes in usage, expensive files, or which projects and models are draining their quota fastest.
github.com
· 2026-08-27
Companies are consuming AI tokens at explosive rates, with Google processing over 3.2 quadrillion tokens monthly—seven times last year's volume—while Uber exhausted its entire 2026 AI budget by April. Industry voices are proposing nutrition-label-style disclosures that let employees compare a prompt's computational cost against its actual value before using it.
fastcompany.com
· 2026-08-27
A proposed technique lets websites detect AI agents through HTTP Accept headers and respond with stripped-down Markdown instead of full HTML pages. This version omits navigation bars, stylesheets, scripts, and layout elements, leaving only the core textual content for automated systems to process.
acceptmarkdown.com
· 2026-08-26
GLM-5.3-Flash is a text-only model with a 400k token context window that scored 57 on the Artificial Analysis Intelligence Index, far above the comparable median of 18. It is priced at $0.15 per 1M input tokens and $0.50 per 1M output tokens, both below the median rates for similar models, with the full benchmark run costing $138.02.
artificialanalysis.ai
· 2026-08-26
GoKawiil