vibehacker
News
GitHub (ollama/ollama) ·

Ollama 0.40.10 adds NVIDIA's Nemotron 3.5 Bolt E4B, a small local vision model with a speculative-decoding draft

Released Oct 23, ollama run nemotron-3.5-bolt:e4b pulls NVIDIA's Gemma 4-based vision-language model for fast local chat, image understanding, and tool use, with a 128K-token context window. It ships with a DFlash draft model for speculative decoding to speed up generation.

More news

View all

Copilot in Visual Studio adds PR exploration via Git agent; BYOM now requires the new Agent (Preview)

In the update posted Oct 22, the Git agent in Visual Studio 2026 can summarize an attached pull request with links to its comments and files (via GitHub or Azure DevOps MCP servers), and NuGet Audit warnings get a fixer that has Copilot patch vulnerable packages. Bring Your Own Model now only works in the Copilot SDK based Agent (Preview), so previously configured Ollama models must be re added…

GitHub Changelog

Study: a coding agent's harness can triple cost per task without improving accuracy

Shandong University researchers held models fixed and ran Claude Code, OpenCode, and the 100 line mini SWE agent on 492 SWE bench Verified tasks, finding statistically equivalent success rates but up to 3x differences in cost per task. On the 45 hardest tasks, switching harnesses flipped results about as often as simply rerunning the same setup (13%), so small single run benchmark gaps say little…

VentureBeat

Spotted something we missed? Start a thread.