AI News: The Most Insane Week So Far This Year!

New Anthropic model that topped benchmarks on release and still leads Artificial Analysis. Claimed 25% cheaper than Fable 5, but Artificial Analysis shows it as the most expensive per task ($3.69). The Megabonk clone looked good but cost about $120 and took 2 hours.

G3Gemini 3.8 FlashGoogleUnverified

Cheap, fast Google model at $0.75/$3.75 per million tokens, scoring 73.7% on the coding benchmark at far lower cost per task than Opus 5 or Fable. Its Megabonk clone looked decent and was generated via Cursor.

AAArtlist AI FlowsArtlistUnverified

Node-based canvas for chaining image, video and voice-over models into reusable workflows, with parallel variations and templates. Included with Artlist Unlimited. Sponsored segment.

Meta model posting top coding benchmark scores (75.4) and third on Artificial Analysis, and free via OpenRouter. The creator's own Megabonk and SVG tests were weak, so he questions the benchmarks' validity.

GAGPT-6 AstraOpenAIUnverified

OpenAI model rolling out to limited organizations then Plus, Pro, Business and Enterprise users. It scored 74.1% on the coding benchmark and near-saturated ARC-AGI-3, and was first on his SVG bench. Its Megabonk clone was the best-looking.

CMChatGPT multiple Google accountsOpenAIUnverified

The ChatGPT app can now connect more than one Google account through its Gmail plugin. The creator showed the manage screen.

GVGemini voice in Gmail, Docs and KeepGoogleUnverified

Conversational Gemini voice features now available in Gmail, Docs and Keep, rolling out to paid AI Plus, Pro and Ultra subscribers.

MRMiniMax real-time video modelMiniMaxUnverified

MiniMax video model generates a 15-second clip in 13 seconds, enabling infinite interactive AI live streams such as infiniteslop.ai and foul.live.

ATAtlasWorld LabsUnverified

World Labs model that takes one or more images plus a camera path and reconstructs explorable 3D scenes with pixel-perfect camera control. In early access, and the creator was impressed.

SOSolarisRunwayUnverified

Real-time interactive video model built on Gen 4.5 that treats clicks and drags as conditioning for the next frame. Early access only, and the creator showed demos.

O2OpenClaw 2.0OpenClawAvailable in EU

Overhauled new version of OpenClaw. The creator has replaced it with Codex and Cursor and hasn't tried it.

Meta transcription model suited to real-time streaming, released September 1st.

MAMAI-Transcribe-2MicrosoftUnverified

Microsoft speech recognition model released September 3rd, billed as the fastest, most accurate and cheapest.

DCDyson CamerajetDysonUnverified

A $500 AI toothbrush with a camera and water jet that aims water between teeth to floss while brushing.