Open-Source Ecosystem Expansion: From Meta-Search to Local LLMs
오픈소스 | Sat Jul 04 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 3 sources
SearXNG meta-search engine, Mistral's Lean 4 proof model, and a guide to running local SOTA LLMs were released.
Analysis
[Mistral AI] open-sourced Leanstral 1.5 under Apache-2.0 license [2]
- 119B total parameters with 6B active parameters architecture
- Saturated miniF2F and solved 587 of 672 problems in PutnamBench
- Achieved state-of-the-art with 87% on FATE-H and 34% on FATE-X
- Available on Hugging Face and via free API
- Discovered 5 previously unknown bugs across 57 open-source repositories
[SearXNG] expanded its free internet meta-search engine open-source project [1]
- Reached 33.2k GitHub stars and 3.1k forks
- Active development history with 9
- 583 commits
- Maintains a separate AI_POLICY document
- Weblate integration for multilingual support
[jamesob/local-llm] published a hardware and configuration guide for running local SOTA LLMs [3]
- $2k budget: 2x RTX 3090 running Qwen3.6-27B and whisper-large-v3 STT
- $40k budget: 4x RTX PRO 6000 configuration with 384GB VRAM
- GPU peer-to-peer communication via c-payne PCIe4 switch
- Serves GLM-5.2-594B at ~80 t/s with 460k context using vLLM
- Provides ready-to-run configuration based on Docker Compose