I've since run the similar AIME25 eval suite against gpt-oss-20b on my own laptop and was surprised to see even that much smaller model solve some of those math problems, again without tools
OPEN SOURCE
-
LangChain and LangGraph 1.0 Alpha Releases Announced
By
–
LangChain & LangGraph 1.0alpha releases Today we are announcing alpha releases of v1.0 for langgraph and langchain, in both Python and JS. LangGraph is a low-level agent orchestration framework, giving developers durable execution and fine-grained control to run complex
-
Serena: Transform Any LLM into Semantic Coding Agent
By
–
Turn any LLM into a semantic coding agent! Semantic Retrieval & Editing Noetic Agent (Serena) is an open-source toolkit that transforms any LLM into a fully-featured coding agent that works directly on your codebase. 100% Open Source
-
Delta Sharing Protocol Wins VLDB2025 Industry Runner-Up Award
By
–
We're excited to share that Delta Sharing: An Open Protocol for Cross-Platform Data Sharing was named Best Paper Industry Runner-Up at #VLDB2025! https://
vldb.org/pvldb/vol18/p5
197-puttaswamy.pdf
… Catch @matei_zaharia
's keynote Thurs (9:15–10:15 BST) and visit our booth or Careers Site for more -
Joyfl v0.1: Minimal Python Interpreter with 342 Lines
By
–
v0.1 : Minimum Viable Interpreter Most notable is that the source code is only ~342 lines — from parser to interpreter, including terminal interaction but excluding comments! ▶︎ You can launch the REPL just by running the `joyfl` script with Python 3.10+. All the samples
-
FastVLM and MobileCLIP2 Vision Language Models Released
By
–
Try it on HuggingFace via desktop & mobile browser! Links to models: FastVLM: https://
shorturl.at/ZWyeN
MobileCLIP2: https://
shorturl.at/z0bch Follow us @futurepedia_io for the freshest AI tools news and updates. -
Chinese food delivery company advances AI through open-source over US tech giants
By
–
A Chinese food delivery company contributes more to advancing AI than US Big Tech thanks to open science and open-source AI The US is cooked if AI scientists don't fight their management to get back to our roots of sharing science rather than keeping it secret and just
-
Joy Programming Language Implementation Updates and Design Elegance
By
–
I should clarify… I believe the *original* Joy design by Manfred von Thun (see link) is the most elegant ever, not this implementation — yet? I'll post implementation updates in the thread below over the next week or so… https://
hypercubed.github.io/joy/html/forth
-joy.html
… -
Microsoft Launches 10B VibeVoice Text-to-Speech Model
By
–
Microsoft just launched a larger 10B parameter version of its VibeVoice text-to-speech model
— The Rundown AI (@TheRundownAI) 2 septembre 2025
Available under MIT license, it generates multi-speaker podcasts (going up to 45 minutes) in minutes
Supports up to 32K contextpic.twitter.com/fO9JOT5RkJMicrosoft just launched a larger 10B parameter version of its VibeVoice text-to-speech model Available under MIT license, it generates multi-speaker podcasts (going up to 45 minutes) in minutes Supports up to 32K context
-
Apple Launches FastVLM and MobileCLIP2 Open Models
By
–
Apple launched two new open models: FastVLM and MobileCLIP2
— The Rundown AI (@TheRundownAI) 2 septembre 2025
The models are up to 85x faster and 3.4x smaller than previous work
Suitable for high-res image processing tasks like OCR, image captioning, visual question answering, and image understandingpic.twitter.com/uG9ObsWACuApple launched two new open models: FastVLM and MobileCLIP2 The models are up to 85x faster and 3.4x smaller than previous work Suitable for high-res image processing tasks like OCR, image captioning, visual question answering, and image understanding