AI — Page 6

Jiang Daxin's flagship pairs a 1M-token window with aggressive pricing, while its own tests still trail top rivals on several coding benchmarks.

The model accepts multiple reference images and turns a three-view character sheet into six scenes, while its open release remains listed as upcoming.

Guillermo Rauch's daily snapshot shows open models dominating volume as Vercel moves closer to the center of the model market.

Micah Hill-Smith and George Cameron's index now separates blocked attempts from agents that recover through a fallback.

Rachel Slank aligned 11 data types and labeled 49,000 craters; the GitHub release includes downstream code but omits the pretraining code.

A Science study reports 0.913 mean AUC and faster reads with AI assistance, while [clinical clearance and patient-outcome evidence remain unestablished](https://ichgcp.net/clinical-trials-registry/NCT07040358).

Its five-year plan pairs $840B in revenue with $856B in compute spending, leaving investors to finance the gap.

CUA-S1-FORMS makes one-pass choices among predefined actions, then hands the plan to Cua Driver for execution.

Grok Voice Transcribe 2.0 reached 2.7% final word error at 0.49 seconds, while its batch result placed fifth in Artificial Analysis testing.

The September 18th release turns Muse into a distribution layer where users can invoke outside services through natural-language requests.

The unreleased interface centralizes agent email, but does not show whether Muse gets an address or reads connected accounts.

The labels say Mikey Shulman's licensed reset carried user feedback and model learnings from earlier systems into Suno's v6 lineup.

SpaceXAI kept pricing at $0.10 per batch hour and $0.20 for streaming while claiming a 2x accuracy gain.

David Fuscus is extending his domain-specific AI play from PR into congressional work, where fluent errors carry statutory consequences.

TypeSafe AI's Jev made 47 Pong decisions in 12 seconds in an Ably demo; Gemini, Claude and GPT made two or three, while still choosing correctly in most runs.

Dnotitia reported a 5.77x FPGA gain, but silicon performance, power and cost remain untested ahead of planned evaluations.

The labs are exploring 20-30 MW deployments as inference demand outruns the timetable for gigawatt campuses.

The integration gives developers typed answers and confidence scores for routing, escalation and tool-call checks inside AI agent workflows.

The model finished 21st overall, but its $1.28 test cost and task-specific wins make aggregate rank only half the buying decision.

Nunchux AI's Muyang Li and collaborators report up to 1.70x faster end-to-end generation without retraining, with MiniMax H3 among four models tested.

The text-output model accepts text, images, audio and video across a 1M-token context window, with function calling and web search for work beyond analysis.

The MiniMax H3-based API generates 5- to 15-second clips with audio, charging $0.02 to $0.075 per output second.

AI engineer Chase Lean's prototype turns the cursor into fish bait, though performance still depends on the machine running it.

The React Native experiment now uses a self-hosted diffusion model, though every booking, payment and order remains simulated.