AIWadmin
Google folds images and audio into one Gemma embedding space
EmbeddingGemma 2 puts text, code, pictures, video and sound on a single 768-dimension vector so retrieval systems stop…
TinyFish drops retrieval fees and bets the cost hides in tokens
The web infrastructure startup made its search and fetch APIs free within rate limits, arguing that page size…
Mistral’s 209M forecasting model tops charts without a GPU
Timestral is Mistral's first time-series family and its largest checkpoint takes first place across three public benchmarks from…
Nvidia frees the GPU scheduler it acquired back in 2024
The full orchestration stack ships under Apache 2.0, giving platform teams a self-hosted alternative to paid Nvidia subscriptions.
MiniMax confirms its stealth model was a Flash release all along
Space Bunny spent ten days at the top of OpenRouter's leaderboard before MiniMax revealed it as the new…
Anthropic grants a slice of equity to a public fund
The Claude maker is placing a slice of its own equity behind a fund it says will pursue…
Google swaps Pytype for a Meta-built type checker
Pyrefly cuts type-checking compute more than 80 percent and incremental rebuild latency by up to 98 percent, replacing…
AWS will cut its EU cloud off the US backbone
On October 24 the European Sovereign Cloud runs for several hours without the global network, and AWS concedes…
Voxel51 rolls robot sensor data and labeling into one stack
Version 3.0 adds time-series multimodal support and an agent with more than 45 MCP tools, plus a free…
Kyutai builds cheap memory into a 2.4B robot model
REKI keeps the last three frames inside its action expert using 32 register tokens, cutting arm oscillation from…
Google ships a framework that retunes an agent at runtime
Harness squared leaves model weights untouched and instead rewrites the prompts, tools and memory around them, lifting GLM-5…
Six agents chasing a train refund turned aggressive fast
Given a hopeless refund claim, agents invented excuses, shared a client's medical history and sent 187 messages in…
New benchmark says agents pick badly on long tasks
Taste-Bench tests whether an agent chooses the better path when outcomes are hidden, and the strongest model managed…
Anthropic drafts conduct code for shopping bots that haggle
The company is consulting law firms on fair negotiation norms, hiring lawyers as research fellows and warning that…
A second German court pins Google to its AI answers
Aachen judges used the Munich reasoning to treat AI Mode output as Google's own speech, and told the…