Ollama: Native Decision Models and Precision Fixes
Ollama shifted decision-model scoring into upstream llama.cpp and added multimodal embedding support, while fixing tool-call precision and model loading reliability. The pattern is less custom code, fewer edge-case failures.
Duration: PT2M11S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-10-12T13:00:31Z
- Audio duration: PT2M11S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Good morning, it's Monday, October 12th, 2026, and this is the Ollama daily briefing.
The lead story is a handoff: decision models now run natively in llama dot C P P, and Ollama is deleting its own scorer to match.
In pull request 18917, merged today, Ollama updates to llama dot C P P build 11521. That build handles the System One endpoint directly and builds each model's prompt internally. That means Ollama drops its custom llama-server scorer and the Clef patch. Related commits align request validation and enable multimodal…
A second theme is API correctness on edge cases. Pull request 18941 fixes tool calls losing precision, where a large I D like 9007199254740993 was rounding down and then persisting rounded in chat history. Pull request 18946 addresses a keep-alive duration so large it overflowed and unloaded models immediately…
Finally, efficiency and Windows reliability. Pull request 18947 reuses verified manifests and loaded runners across generate, chat, embed, and System One, avoiding repeat lookups. A new opt-in parallel setting lets hybrid and recurrent architectures use multiple slots instead of one. On Windows, one fix resolves…
What's next: test decision-model prompts…
Nearby episodes from Ollama
- Weekly Recap - Decision Models Go Native, Embeddings Go Multimodal
- GPU Discovery Fallback and Docs Cleanup
- OpenAI Parity and Reasoning Fixes
- Legacy Model Cleanup and Agent Launcher Polish
- Memory Accuracy, Tool Fixes, and Simpler Onboarding
- Multimodal Embeddings and Cloud Convergence
- MLX Performance Push and Tool-Call Reliability
- Tokenizer Parity and RC-Aware Updates