Announced Tue · reported Tue
OpenAI announced Jalapeño, a custom inference chip that offers faster, more power-efficient inference with higher throughput and lower latency for modern models.
DecoderAnnounced Wed · reported Thu
Google DeepMind announces Gemini 3.5 Transcribe, a new speech-to-text model for real-time transcription, available via the Gemini API in Google AI Studio and Gemini Enterprise Agent Platform. The model offers smart transcription, function calling, custom vocabulary, multilingual support, and multi-speaker identification. It is accessible through streaming (Live API with gemini-3.5-transcribe-live) and pre-recorded audio (Interactions API with gemini-3.5-transcribe) APIs.
Announced Tue · reported Tue
OpenAI introduces the Admin plugin for ChatGPT Work and Codex, enabling workspace usage analysis, member and permission management, limit adjustments, and handling of admin requests.
24 August
NVIDIA announced that its Groq 3 LPX inference accelerator is in full production, delivering 3,400 output tokens per second on Gemma 4 31B for 100,000-token long-context use cases, 4x faster than alternatives. The system is part of the Vera Rubin platform and is being adopted by partners like SpaceXAI, CoreWeave, and Nebius.
25 August
Anthropic announced that Claude's memory now works across both chat and Claude Cowork, allowing users to see, edit, and delete memories. Users can also enable memory for sensitive topics like health or beliefs, while default memory excludes sensitive information.
OpenAI announced Jalapeño, a custom inference chip that offers faster, more power-efficient inference with higher throughput and lower latency for modern models.
26 August
Anthropic announced that Claude now has a built-in browser in the Claude Cowork desktop app, enabling Claude to navigate websites, read pages, and fill forms without requiring the user's own browser. The feature is rolling out to Pro, Max, and Team plans on desktop, with Enterprise admin control available. The built-in browser is separate from the user's browser and does not access user tabs, bookmarks, or passwords.
OpenAI shares findings from a security incident involving Hugging Face and outlines steps to strengthen AI model security, monitoring, and alignment.
27 August
Anthropic is previewing the Model Hardware Standard (MHS), a shared specification enabling AI agents to safely operate physical devices like microscopes and robotic arms, reducing integration time from weeks to hours or minutes.
Gemini Omni 1.1 Flash adds scene extension, keyframe interpolation, 4K upscaling, and 360p drafting.
28 August
OpenAI ending model contract with Cursor after SpaceX acquisition
29 August
30 August
A local-only version of Perplexity Computer for on-device AI workflows.
Perplexity25 AugOpenAI enforced its safety policy by banning accounts originating from Russia that were leveraging AI to promote a fabricated think tank and a sovereignty index that favored Russia and disparaged the West.
Google DeepMind announces the world's first double-blind evaluation of a proprietary frontier AI model, using cryptographic environments to prevent benchmark contamination. The pilot tests a Gemini Flash Lite model with external partners against confidential benchmarks in a privacy-preserving environment, enhancing evaluation integrity and trust in model assessments.
Plugin marketplaces can now automatically update installed plugins via enterprise managed settings.
