Ollama v0.23.4 brings vision support to opencode, fixes Claude tool formatting
Run vision models locally with image inputs – plus a Claude tool result fix.
Ollama, the wildly popular open-source tool for running local LLMs (171K GitHub stars), shipped v0.23.4 with a notable upgrade: the `opencode` launch command now supports vision models that can accept image inputs. This means developers can use local models like LLaVA or other multimodal LLMs to process images directly from the command line, without sending data to external APIs. The release also resolves a formatting bug where Claude tool results would break when referencing local image paths, improving reliability for those using Claude models with local media.
The update follows the v0.23.3 release and includes three confirmed commits from the main branch. Community reaction was positive, with 5 rocket emojis, 4 heart reactions, and 3 thumbs-up. Assets are available for download across platforms. For engineers running local AI pipelines, this small version bump unlocks a meaningful new capability – vision inference without cloud dependencies – and polishes an existing integration pain point. As local models grow more capable, Ollama’s rapid iteration keeps it central to the self-hosted AI ecosystem.
- Ollama v0.23.4 adds vision model support to the `opencode` launch command, enabling image inputs locally.
- Fixes formatting of Claude tool results when using local image paths, improving reliability.
- Release includes 3 commits since v0.23.3 and received strong community engagement (5 rocket reactions).
Why It Matters
Local vision models become more accessible, and Claude tool integration gets a critical fix for self-hosted workflows.