Thinking Machines released Inkling 2 Small, a 594B MoE with 17B active parameters, scoring 31 on Artificial Analysis.

Google released Gemini Omni 1.2 Flash, a video generation model with native 15-second clips at $0.07 per second.

Baidu has released ERNIE 5.5 Flash Base, the raw pre-training checkpoint of its new sparse Mixture-of-Experts language model.

OpenAI’s GPT-Image-2.6 Flare and Sunburst bring 50% faster image generation, GIFs, SVG exports, and complexity control to ...

Together AI’s Together Compiler uses AI agents to continually recompile inference, beating InferenceX baselines by up to 2.6x ...

AI’s TwIL-8B is an 8.8B open-weights reasoning model scoring 98.2 on MATH-500 and running locally in 4.98 GiB.

Ricursive’s Ripple-65B Preview is a 65B dense open-weight model scoring 80.5 on Terminal-Bench 2.1, released under Apache 2.0 ...

Anthropic FlashProve generates zero-knowledge proofs for 400B Llama 4 Maverick inference at 110 tokens/second on 1 H100 GPU.

NVIDIA shows 3 ways to speed up MoE inference on DGX Spark, lifting Ling-3.0-flash GSM8K decode throughput 42.5%.

Google Agile Deliberation helps define subjective visual concepts, lifting VLM classifier F1 by 10.5 points over zero-shot ...

Tencent WeVisOne-VL is an open 2B/4B vision language model scoring 71.3 F1 on Dense200 without a separate SAM.

Size: Soniox STT v6 is a proprietary hosted model, no parameter count published. Runs on: Soniox cloud API only (WebSocket ...