⚡️A 9B embodied AI model beats GPT-6 Astra on 6 of 8 spatial grounding benchmarks. ACE Robotics' ACE-Brain-0.6 is fine-tuned from Qwen3.5-9B using ~1M spatial grounding samples with verifiable visual evidence. > 74.53 average across 7 public benchmarks, vs. 73.68 for Astra > 94.13 on ACE’s RobotGrounding benchmark, vs. 91.10 for Astra Beyond locating objects, the model can identify when a target is missing — a critical capability for robots that need to know when NOT to act. The data scaling results are just as interesting: > expanding training data from 260K to ~1M samples raised the average score from 69.3 to 74.5. > Simply adding more of the same data types improved scores by less than 1 point. New scenes, spatial relationships and diverse supervision drove the bigger gains.

$23.9B of stablecoins is deposited across 37 DeFi venues Aave V3 and Morpho Blue account for 64.3% of the total, or ~$15.4B Stablecoin growth directly expands the addressable market for onchain lending

A free 9B model, fine-tuned by @coinbase for under $100, beat Opus 4.5 at fraud detection. Acurast runs that same model (Qwen3.5-9B) on a phone. How fast is it? What does it cost? How does it scale to 100+ phones? Read the full test results and deploy it yourself 👇

Introducing SenseNova-Loop-9B-Preview: a 9B open-weight multimodal model with 256K context for document understanding, coding, and agent workflows. 📜 Apache-2.0 License. 🤖 • Leads six of eight tasks in the team’s reported comparison: Scores 71.38 on BIRD-Dev, 66.89 on WorkBuddy Bench, and 58.87 on AIDABench. • A deeper path without duplicate weights: Its 24-layer middle block runs twice with shared weights, producing 56 layer executions from 32 distinct decoder layers. • Trained for long, multimodal work: Training recipes combine text, images, code, and tool-use trajectories at sequence lengths up to 262K tokens. • Acceleration and serving paths are available: MTP is supported now, with documented Transformers, LightLLM, SGLang, and vLLM deployment paths. Note: This is a Preview release; the technical report is still forthcoming. Examples created with Raccoon Work 👇

Built on a 9B parameter architecture with hybrid attention for efficiency. The ternary 2-bit quantization keeps it tiny while preserving multimodal understanding. GGUF format means it plays nice with llama.cpp across CUDA and Metal backends.

Meet Ternary-Bonsai-2-9B-GGUF: a 9B image-text-to-text model squeezed into 2-bit ternary weights. Runs on llama.cpp with CUDA and Metal, so you can chat about images right on your laptop or phone. Small but mighty!

Small vision-language models (9B total, 2B active MoE): image + text in, 32K context, optional reasoning. Use /caption {"length":"normal"} for captions Both are $0.30/M input and $2.50/M output tokens See more: