Ollama: Faster Concurrent Inference and Unified Desktop Launch
Concurrent inference gets faster with batched speculative verification and more accurate memory limits for firmware-reserved iGPUs, while ChatGPT and Claude Desktop launch flows converge on shared settings. Chat-path efficiency and Windows reliability fixes round out the day.
Duration: PT2M19S
Episode overview
This episode is a short developer briefing from Ollama.
It explains recent repository work in plain language.
- Show: Ollama
- Published: 2026-11-12T14:01:13Z
- Audio duration: PT2M19S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
Hello, this is your Ollama briefing for Thursday, November 12th, 2026.
The through-line today is throughput and trust: making concurrent local runs faster without overpromising memory, while making desktop connections behave consistently.
First, inference performance. Change 19663 keeps speculative drafting active when multiple generation requests overlap, verifying proposed tokens in batches instead of disabling the speedup. That should help clients running side-by-side finish sooner, at the cost of more temporary GPU memory and possible first-token…
Second, desktop launch is converging. Changes 19658 and 19660 make the ChatGPT button and command-line launch share the same model list, defaults, and reconnect logic, so settings choices persist instead of being replaced by a single-model selector. In parallel, 19657 and 19656 move the Claude Desktop gateway onto…
Finally, chat-path hygiene. Changes 19665 and 19666 replace per-message maps with typed request structs and switch streamed tool-call accumulation to a string builder, cutting reflection and quadratic copying while keeping output byte-identical. And 19668 fixes a Windows process check that mistook a byte count for a…
Expect…
Nearby episodes from Ollama
- Tool Schema Fidelity, Desktop Launch, and Download Safety
- Tool Reliability and Mac Inference Tuning
- MLX Goes Parallel
- VRAM Fixes and Launch Polish
- Stricter Model Imports and CLI Output Fixes
- Weekly Recap - MLX Model Expansion and Launch Polish
- Ripple on MLX and Validation Hardening
- MLX Memory Cuts and Tool-Call Fixes