Anthropic’s quarterly revenue figures obtained by the Financial Times triggered investor panic despite strong growth and very high reported sales. The quarter reportedly brought in $16+ billion revenue with 46% quarter-over-quarter growth, after earlier growth of 145% the previous quarter. As a result, the article argues expectations for sustained exponential growth are starting to look like they may not be met, making an AI “bubble” closer to popping.
Meta launched Muse Spark Open, an 816B open-weight AI model with 42B active parameters, and made it available via multiple model APIs and major model hubs under the Muse Model License. The model uses a 1 million token window. The license adds limits tied to developers’ revenue and requires UI attribution (“Built with Muse Spark Open”), while Meta’s benchmarks claim it performs at or above several top Chinese open-weight models but trails them on specific coding tests like Terminal-Bench 2.1.
Anthropic’s private investor briefing leaked within minutes and the leaked figures were used to reassess the company’s AI-related revenue outlook. Third-quarter revenues were reported as $16.8bn, with annualized exit revenue of $76bn and net revenue retention around 900%. The result is a market shift toward slower-growth expectations and uncertainty over whether Anthropic can hit $60bn full-year revenue as it prepares to go public via an S-1 filing.
Nvidia announced new Nemotron 3.5 models, including a 120B vision-language model and two smaller Bolt models meant for local use. Nemotron 3.5 Bolt 74B-A9B is quantized to fit in 24GB of memory. The launches add multimodal enterprise agent capability plus downloadable local models on Hugging Face under an OpenMDW license for commercial use and modification.
Jev released an improved Jev 1.14.0 plus experimental “danger zone” API features aimed at less-reliable, creativity-focused use cases. The experimental image and longer-context options are enabled via headers named TypeSafe-Danger-Zone-Images: true and TypeSafe-Danger-Zone-Long-Context: true. Developers can now opt in to composable image support, add a new box type, and use longer context lengths, while production use is discouraged for these experimental capabilities.
Apertus released its Apertus 1.5 models, extending the prior 8B and 70B Aptetus 1.0 generation with image/audio understanding, an optional thinking mode, longer context, improved instruction-following, and better tool use. The updated 8B and 70B weights were posted to Hugging Face on July 24, 2026. The release adds multimodal token-based processing, a 262,144-token context window, and new post-training/evaluation materials that improve reasoning, coding, and multimodal document/chart performance.
Anthropic reported about $16.8 billion in third-quarter 2026 revenue ahead of its planned IPO, after presenting the figures to prospective investors in San Francisco. The third-quarter revenue figure is up roughly 46% from the previous quarter. Investors are being pitched on high adjusted profitability and cash generation, while the company’s use of “adjusted” metrics keeps some major costs and partner cuts out of the headline margins.
TypeSafe AI released Jev 1.14, an update to its System One typed decision model that is live under the jev-preview alias with new experimental capabilities. The default context is 64k tokens per request and 32k per question, with an opt-in header that raises this to 262k tokens. Developers can now enable experimental image input and a native Box bounding-box question type, but TypeSafe says none of the three new danger-zone features are ready for production.
SpaceXAI released Grok Voice 3.0, a speech-to-speech voice agent model exposed via a realtime WebSocket endpoint. It scored 86.9 to rank #1 on the Artificial Analysis Speech-to-Speech Index. Developers keep the same endpoint/events and $0.08 per minute price while the grok-voice-latest alias updates automatically, with Grok Voice 3.0 available only via hosted API (no weights).
Every AI story that matters,
in your inbox by 8am.
TLDRocket reads all relevant sources, removes duplicate coverage, and summarises the
day in two minutes. Follow companies and topics for alerts, or get the
briefing in Slack. Free, no spam, unsubscribe anytime.
Reading TLDRocket needs no cookies, and the readership counts we rely on come from
our own cookieless analytics. Google Analytics is the exception: it sets cookies and
reports to Google, so it stays switched off until you allow it. You can change your
mind any time from “Cookie settings” in the footer.
Strictly necessary
Session security and form protection (tldrocket-session,
XSRF-TOKEN, 2 hours). The site cannot work without them,
so they need no consent.
Always on
Google Analytics 4 (_ga,
_ga_<id>, up to 2 years). Measures which
stories and sections readers use. Google acts as a third-party processor and may
store the data outside the EU. No advertising, no profiling, no data sold.
