nano
10 stories and discussions about nano, aggregated from every source we track.
Targeting AI ran autonomously on non-frontier models.
The Nano Banana 2 Lite MCP server from July, updated: FastMCP is now MCPServer, google-genai 1.x gets a 400 from the Interactions API, and one server now runs in Claude Code, Codex and Antigravity CLI.
The Kiro edition of the Nano Banana 2 Lite MCP server, updated: FastMCP is now MCPServer, google-genai 1.x gets a 400 from the Interactions API, and Kiro CLI 3 needs a permissions rule before it will call the tools headless.
Alibaba Group's AI division has released Qwen 2.1 Image, an ultra-lightweight image generation model with just 7B parameters, able to run on local hardware like the RTX 3090, putting out high-quality images in less…
FPGA example designs that were created using hotstate and the hotc compiler (for more information on the Hotstate Machine see: https://hotwright.com/) - philtomson/hotc_examples
FPGA example designs that were created using hotstate and the hotc compiler (for more information on the Hotstate Machine see: https://hotwright.com/) - philtomson/hotc_examples
PanTilt 0.9.0 - desk visual tracking with journal + optional Nano/PCA9685 neck. Same ORT/OpenVINO loop family as edgeinfer-eval. - olesha-ai/pan-tilt-ai-tracker
<p>This particular model release is interesting because not only is it open weights[0], but they also released most of the pretraining and posttraining data along with the code and libraries for the RL environments (much like Olmo3, which came out recently). It is also trained from scratch and architecturally different from the existing Qwen and DeepSeek templates. It is frontier performance for its weight class. Two larger model releases are to follow soon.</p> <p>[0] Not exactly open source, though—it's Nvidia Open Model License.</p>