Ollama v0.40.17-rc0: MLX concurrent generation, mmap on Vulkan integrated GPUs, thinking metadata in OpenCode config
Announced November 10, 2026
Verified (fact)
A release candidate for Ollama v0.40.17 is out. The MLX backend now supports concurrent generation, Vulkan integrated GPUs now use mmap, and OpenCode launch configuration now writes thinking metadata. It also bumps llama.cpp (b12257) and MLX/MLX-C versions.
Why it matters (interpretation)
It makes local serving that handles multiple requests concurrently on Apple Silicon easier.
Sources
- Official v0.40.17 Ollama
More about Ollama
-
Inkling 2 Mini follow-up: Ollama v0.40.17 runs it out of the box ('79.2% on SWE-Bench Verified'), adds concurrent requests on Apple silicon
Ollama's v0.40.17 release (tagged rc4) supports Thinking Machines Lab's Inkling 2 Mini. It can be run with a single command, `ollama run...
-
Thinking Machines Lab releases looped open-weight model Inkling 2 Mini: 26B total, 5.4B active, runs on a Mac in ~20GB
Thinking Machines Lab released Inkling 2 Mini, an open-weight model presented on the Hugging Face blog and in the Ollama v0.40.17...
-
Ollama v0.40.16-rc0 fixes Claude Code routing errors, cuts Qwen3.8 Flash Next memory use on MLX, adds CUDA 13.5 for ARM64 Windows
Ollama released a v0.40.16 release candidate. It fixes routing and deprecated-model errors when running Claude Code, reduces memory use...
-
Ollama v0.40.15 released: adds support for Xiaomi's MiMo V2.6 (9B) and the 'AutoTrust' decision model family
The v0.40.15 stable release can run Xiaomi MiMo V2.6, a 9B model for coding and agent tasks. The newly added AutoTrust is a decision...
-
Ollama v0.40.14 applies model output limits when launching Claude Code and respects sampler defaults in OpenAI-compatible API; v0.40.15-rc0 adds MiMo V2.6
In v0.40.14, 'ollama launch claude' applies per-model output token limits so cloud models no longer hit the 128k default cap. The...