Ollama v0.40.16 Adds Ripple Preview 65B, Fixes API Truncation, and Cuts Mac Memory Use
What changed
Ollama released version v0.40.16, introducing support for Ripple Preview, a 65B dense model built by Ricursive Intelligence targeting agentic coding tasks. The model can be run locally using ollama run ripple-preview.
The update includes several hardware and API modifications:
- Memory Reduction on Apple Silicon: Running Qwen3.8 Flash Next on Mac hardware now consumes approximately 27 GiB less memory.
- API Behavior Fixes: The
/v1/responsesendpoint now respects thetruncationparameter and no longer truncates response payloads silently by default. Meanwhile,/v1/modelsreturns an empty array[]instead ofnullwhen no local models are installed. - Tooling and Integration Updates: Executing
ollama launch claudenow reliably forwards Claude Code requests to the chosen model. Additionally,ollama launchdisplays an explanatory message whenever a selected cloud model has been retired. - Platform & Runtime Updates: Added CUDA 13.5 support for Windows on ARM, alongside an updated build of the underlying
llama.cppengine.
Why it matters
For developers executing agentic coding workflows locally, Ricursive Intelligence’s 65B dense model brings another specialized parameter option to the Ollama library.
The API modifications resolve subtle client-side bugs for integrations relying on OpenAI-compatible endpoints. By ending silent default truncation on /v1/responses, Ollama avoids unexpected text cuts during long code generation tasks. Correcting /v1/models to return an empty list rather than null prevents type error exceptions in client SDKs parsing missing local models.
Memory footprint cuts of 27 GiB for Qwen3.8 Flash Next on Apple silicon significantly reduce VRAM pressure, making high-parameter models easier to host alongside developer tools. Furthermore, enforcing consistent model targets in ollama launch claude provides better routing reliability for users working with workflows like Claude Code.
What to do next
To update to the latest release and test the new model:
- Upgrade Ollama to version v0.40.16 via your installation manager.
- Execute
ollama run ripple-previewto download and evaluate the 65B agentic coding model. - Verify downstream API integrations using
/v1/responsesto ensure any custom truncation rules align with the new explicit parameter behavior.
Caveats: Because Ripple Preview is a 65B dense model, hardware with significant memory capacity is required to achieve playable inference speeds. Early preview status means performance and agent behavior may evolve in future model revisions.
Sources
Share Article
If this article helped you, please share it with others!