Asif Razzaq

Harness-1 is a 20B search agent reaching 0.730 average curated recall across eight benchmarks, behind only Opus-4.6.

Microsoft AI releases MAI-Transcribe-1.5, a speech-to-text model covering 43 languages with faster long-form audio transcription.

Xiaomi MiMo-V2.5-Pro-UltraSpeed decodes past 1000 tokens per second on commodity GPUs using FP4 quantization and DFlash speculative decoding.

Harvard and Perplexity find AI agents do 26 minutes of autonomous work per session, versus Search's 33 seconds

Always Active Necessary cookies are absolutely essential for the website to function properly.

Anthropic releases Claude Fable 5 and Mythos 5, two Mythos-class models sharing one model, differing only by safeguards.

Google AI releases DiffusionGemma, a 26B MoE open text diffusion model generating 256-token blocks in parallel, up to 4x faster.

Cohere released North Mini Code, an open-weight 30B mixture-of-experts coding model activating 3B parameters for agentic engineering

Zyphra released Zamba2-VL, open hybrid SSM–Transformer VLMs at 1.2B, 2.7B, and 7B with lower time-to-first-token than Transformers.

Moonshot AI launches Kimi Work: a local desktop agent pairing a 300-sub-agent Agent Swarm with WebBridge browser automation.