Ollama Release Notes

What's Changed

Enabled flash attention on older NVIDIA GPUs (compute capability 6.x)

iGPU can now offload vision models with padding to fit available memory

Fixed structured output for thinking models when thinking is disabled

Hardened GGUF model creation

ollama launch for Claude Code now disables telemetry by default

Fixed loading models on paths with non-UTF-8 characters

Updated the MLX and llama.cpp engines

New Contributors

@kevinpark1217 made their first contribution in #16949

Full Changelog: v0.31.1...v0.31.2

llm: allow iGPU mmproj offload with fit padding

llama.cpp's fit pass sizes text-model placement before the multimodal projector is loaded. Ollama had been avoiding that risk on non-Metal iGPUs by disabling projector offload entirely, which forces CLIP onto CPU on GB10 and Strix Halo even when the projector has ample memory available.

Let integrated GPUs use the same projector-memory check as other GPUs. When projector offload is enabled, add the estimated projector memory plus the existing 1 GiB headroom to Ollama-owned LLAMA_ARG_FIT_TARGET so fit leaves space for the later projector allocation. If Ollama/device setup already supplied a fit target, add the projector pad to it. If the user set LLAMA_ARG_FIT_TARGET explicitly, leave it exactly as provided.

Fixes #16419

review comments

Faster Gemma 4 on Apple Silicon

Gemma 4 is now significantly faster in Ollama on Apple Silicon, generating tokens nearly 90% faster on average across a coding-agent benchmark by leveraging multi-token prediction (MTP). Ollama auto-tunes how many tokens to draft as it runs, so the speedup is on by default, requires no configuration, and does not change the model's output.

What's Changed

• Tightened Gemma 4 MoE model loading in the MLX engine
• Updated the MLX engine to the latest version, including a new small-batch matmul kernel
• Updated the underlying llama.cpp engine to build 9840
• Improved Gemma 4 multi-token prediction (MTP) performance

Full Changelog: v0.30.12...v0.31.1

What's Changed

launch: add thinking capability detection to opencode by @hoyyeva in #15434
launch: auto-install Claude Code by @hoyyeva in #16802
launch: auto-install opencode when missing by @hoyyeva in #16806
discover: fix inverted iGPU/dGPU Vulkan classification on Windows hybrid graphics by @Sahil170595 in #16669
mlxrunner: unify and tune speculative decoding by @jessegross in #16791
launch/codex: detect model drift when Codex App UI switches by @BruceMacD in #16864
llama: add sm_86 architecture to cuda_v13_windows preset by @anishesg in #16834
llm: size mmproj offload by projector memory by @dhiltgen in #16866
docs: document max think level by @ParthSareen in #16877
llm: preserve generation headroom for shifted prompts by @ParthSareen in #16856
llama: default qwen2.5vl window attention metadata by @dhiltgen in #16868
llm: use host Vulkan loader on Windows by @dhiltgen in #16869
mlx: update and fix CUDA JIT packaging by @dhiltgen in #16871
llm: fix ollama ps double-counting mmap'd weights on partial offload by @discobot in #16709
docs: redesign docs landing and integrations overview by @hoyyeva in #16807
server: align generate with native chat templates by @dhiltgen in #16878
jetson: add CC 87 for CUDA v13 by @dhiltgen in #16628
llama.cpp version update by @dhiltgen in #16548

New Contributors

@Sahil170595 made their first contribution in #16669
@anishesg made their first contribution in #16834
@discobot made their first contribution in #16709

Full Changelog: v0.30.10...v0.30.11-rc0