Menlo Research (@menloresearch) / X

Menlo Research

764 posts

Menlo Research

@menloresearch

We deploy robots to solve real world problems. Get in touch: menlo.ai/talk

Joined May 2024

Menlo Research
@menloresearch
Oct 14, 2024
🍓 Ichigo-llama3.1: Local Real-Time Voice AI We bring 2 key improvements to Ichigo: - It can talk back (Yes!) - It recognizes when it can't comprehend input You can now run this little strawberry on your device! Demo on a single @nvidia 3090 GPU. 1/10
00:00
Menlo Research
@menloresearch
Jun 17, 2025
Meet Jan-nano, a 4B model that outscores DeepSeek-v3-671B using MCP. It's built on Qwen3-4B with DAPO fine-tuning, it handles: - real-time web search - deep research Model + GGUF: huggingface.co/collections/Me… To run it locally: - Install Jan Beta: jan.ai/docs/desktop/b… - Download
00:00
Menlo Research
@menloresearch
Jul 22, 2025
Introducing Lucy: 1.7B model that Google for you It's an agentic‑search model that can even run on your phone. - Agentic search on tap - Lucy calls tools (<think></think>‑aware) - Fits in your pocket - runs on CPU or mobile Under the hood: - Built on @Alibaba_Qwen's Qwen3‑1.7B
00:00
Menlo Research
@menloresearch
Feb 21, 2025
AlphaMaze: Teaching LLMs to think visually We're excited to share our findings—now live in a blog, paper, and open-source release. Language models are good at words—but what about visual-spatial reasoning? AlphaMaze is a decoder-only model trained to think visually by solving
GIF
Menlo Research
@menloresearch
Oct 21, 2024
Meet 🍓 Mini Ichigo: Ichigo-llama3.5-3B-s Mini Ichigo is a local real-time voice AI built on @AIatMeta's Llama3.5-3B. It scored 59.61 on MMLU, close to larger models like llama3.5-instruct-8b (69.40), and held up well in AudioBench with 2.58 on Open-hermes and 2.07 on Alpaca.
Menlo Research
@menloresearch
Sep 18, 2024
Control AI with your voice. The upcoming Llama3-s checkpoint introduces voice-based function calling. It's an early-fusion model built to give @AIatMeta's Llama 3.1 model listening capabilities. Coming soon to @huggingface & @Gradio.
00:00
Menlo Research
@menloresearch
Sep 27, 2024
Our homebrewed early-fusion speech model now has a name and a voice. Say hi to 🍓 Ichigo, the local real-time voice AI. Next up: You can try Ichigo soon on @huggingface & @Gradio.
00:00
Menlo Research
@menloresearch
Apr 16, 2025
ReZero: A small model that learns to search - it never gives up 🔥 ReZero trains with synthetic search engines that force the model to retry, refine, and persist until it finds a better answer (never give up 💪). It's built on Meta's Llama 3.2B. Instead of optimizing for recall
00:00
Menlo Research
@menloresearch
Nov 11, 2024
🍓 Ichigo is smarter now with v0.4 - MMLU score up to 64.66 - Better at picking out voices in noisy settings - Tracks complex, multi-turn conversations better Links: - GitHub: <github.com/homebrewltd/ic…> - Live demo: ichigo.homebrew.ltd - Model weights:
Menlo Research
@menloresearch
Dec 23, 2024
IchigoWhisper: Speech recognition for low-resource languages. We're building an ASR for low-resource languages, including Southeast Asian languages like Vietnamese. Low-resource languages face challenges like limited datasets and high word error rates. Existing models,
Menlo Research
@menloresearch
Mar 12, 2025
We made progress on a robotic hand manipulation model, PoseLess. PoseLess reproduces 2D images of hands directly in 3D simulation, without complex inverse kinematics. Here's how 👇 PoseLess is a depth-free vision-to-joint model that eliminates the intermediate pose estimation
Menlo Research
@menloresearch
Dec 20, 2024
We're training a speech model with the speechless method. First, training speech models is tough - collecting Q&A speech data is expensive, slow, and doesn't scale. Traditional pipelines rely on generating audio (TTS -> ASR), adding unnecessary complexity. With the Speechless
Menlo Research
@menloresearch
Mar 25, 2025
Homebrew (Jan) is now Menlo Research. Same mission, sharper focus: Menlo is an open R&D lab building the brain for robots. The work has grown. So has the team. Robotic agents, hardware — it made sense to give it all one name. Same values: - Jan, Cortex, and models stay open -
Menlo Research
@menloresearch
Aug 23, 2024
Llama3.1 just got ears with llama3s! 🦙👂 We're teaching Llama3.1 to listen - an open, ongoing experiment with @AIatMeta's llama3. This v0.2 can understand human speech, but it's in its early days with limitations/bugs.
00:00