🤖 Gemini 3.7 Flash ships with 50% API price cut 🤖
Google's rapid rollout pressures competitors on pricing and speed.

Just three weeks after Gemini 3.6 Flash, Google has launched Gemini 3.7 Flash, delivering major upgrades for coding, web development, knowledge work, and long-document handling. The release comes with aggressive introductory pricing — API costs are temporarily halved to $0.75 per million input tokens and $3.75 per million output tokens through year-end, applying direct competitive pressure on OpenAI and Anthropic's comparable mid-tier offerings. This rapid release cadence signals Google's intent to own the mid-tier model segment, where most enterprise workloads run. The combination of performance gains and cost cuts makes Gemini 3.7 Flash one of the most consequential model releases of the summer, forcing rivals to respond on both capability and pricing.

🧠 Z.ai's GLM-5.3 tops open-source coding benchmarks 🧠
Z.ai's open-source rollout claims top coding scores with emergent cyber capabilities.

Chinese AI lab Z.ai released GLM-5.3, claiming it is the strongest open-source coding model currently available, with top scores on several cyber benchmarks. All performance gains came from scaled post-training on a 743B-parameter base model, leveraging more environments, diverse tasks, and increased compute. Open weights are expected within two weeks. The model demonstrates emergent capabilities in cybersecurity-adjacent tasks, raising both excitement and safety questions — a signal of competitive pressure from China's accelerating reinforcement-learning ecosystem. Reports suggest US data companies are selling training data to Chinese labs, fueling this pace. GLM-5.3 positions Z.ai as a serious contender at the open-source frontier, with a release cycle measured in days rather than months.

💻 OpenAI rolls out Computer History for Mac 💻
OpenAI's new on-device integration tracks desktop activity for persistent ChatGPT context.

OpenAI launched Computer History, replacing its earlier Chronicle preview, as an opt-in feature for Mac users. The tool tracks clicks, typing, app switches, and accessibility events locally, then compiles them into summaries that ChatGPT and Codex can access to understand what users have been working on. All processing and summarization happen on-device, addressing privacy concerns. This is a major step toward persistent, context-aware AI assistants that operate across an entire desktop workflow — an integration play that could reshape how knowledge workers use AI daily. The move positions OpenAI to compete with similar ambient-computing efforts from Apple and Google, expanding AI beyond chat into continuous desktop assistance.

INTERESTING TO KNOW

🛡️ Amodei fires back at AI regulation critics 🛡️

Anthropic CEO Dario Amodei engaged in a rare public exchange on X, pushing back against critics who say his safety messaging has hurt AI's public reputation. Amodei argued the only way AI regains trust is by “actually curing cancer,” pointing to early biology results as more persuasive than marketing. He framed the regulation debate as a “false choice,” saying Anthropic's proposals would slow big labs while sparing small ones — a regulation distinction around access and pricing that could shape future AI policy.

🏢 OpenAI loses key executives ahead of IPO 🏢

As OpenAI accelerates its IPO rollout, a wave of senior executive departures is raising investor concerns. Chief Revenue Officer Denise Dresser, who joined from Salesforce less than a year ago, is leaving and will be replaced by Dali Rajic from cybersecurity firm Wiz. Executive Brad Lightcap also departed earlier this week to start a new venture. The rapid leadership churn at a critical moment puts pressure on the new team to stabilize operations ahead of what could be one of the largest tech IPOs in history.