Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Ollama v0.40.14 fixes Claude/Codex launch limits, cloud search decoding, and OpenAI sampler defaults
ollama tagged v0.40.14. `ollama launch claude` now respects the model’s output-token limit for Claude Code instead of the 128k default; cloud web-search `/api/chat` responses decode correctly when compressed; OpenAI-compatible endpoints keep packaged sampler defaults when clients omit temperature/top_p; `/v1/responses` keeps readable input including prior refusals; the app retries ChatGPT disconnects; `ollama launch codex` verifies ChatGPT routing before configure; and MLX top-k sampling is faster for large vocabularies.
- Ollama v0.40.12-rc0 adds MLX DFlash2 and Gemma vision detail fixes
Ollama published v0.40.12-rc0 on GitHub. Changes include preserving small-image detail for compact Gemma 4 vision towers on MLX, MLX prompt-speed bumps for gemma/flint/laguna, MLX DFlash2 support, and a llama.cpp bump to b11901.
- Ollama v0.40.11-rc0 tightens Qwen3 Coder tags, OpenCode context, and Phonon voice
Ollama tags v0.40.11-rc0 with fixes that preserve literal function closing tags in Qwen3 Coder parameters, use allocated context for OpenCode, allow empty state for vllm-d3 decisions, bump llama.cpp, and expand Phonon voice catalog and symbol spelling in mlxrunner.
- Ollama v0.40.11 adds on-device speech on macOS and d3 decision models
Ollama v0.40.11 lets `ollama run` speak replies on macOS as audio is generated and exposes an OpenAI-compatible `/v1/audio/speech` endpoint, and ships d3 decision models from vLLM Semantic Router for classify / yes-no / score tasks across 2B–27B sizes.
- Ollama publishes v0.40.9-rc2 with a llama.cpp update
Ollama tagged v0.40.9-rc2 on GitHub. The release notes list a llama.cpp update versus v0.40.8 as the main change in this release candidate.
- Ollama adds Phonon speech generation on Apple Silicon with streaming audio playback
Ollama's v0.40.9-rc3 release notes say Speech in Ollama can generate speech with Phonon on Apple Silicon: ollama run can play text or a chat model's reply aloud as the audio is generated, and apps can stream audio through the OpenAI-compatible /v1/audio/speech endpoint.
- Ollama 0.40.9 adds GA Gradium Phonon speech checkpoints
Ollama released v0.40.9 with support for Gradium's GA Phonon speech checkpoints, replacing the prerelease path, including default-voice import and speech conditioning, and updating the speech docs and tests.
- Ollama v0.40.8 adds Flint model support to its MLX backend
Ollama's v0.40.8 release adds support for Flint models on the MLX backend and preserves the source tool JSON alongside decoded API values, so Flint keeps numeric types and schema order in prompts. Existing API marshaling is unchanged.
- Ollama says it has scaled its US compute footprint with Crusoe
Ollama says it has spent months expanding its compute in the US with Crusoe, to offer day-zero models and full zero data retention.
- Ollama v0.40.7-rc0 adds PPLX Decider v1.2 support to its MLX backend and lets Responses requests set effort on non-thinking models
Ollama's v0.40.7-rc0 pre-release adds support for PPLX Decider v1.2 in the MLX backend. Its OpenAI-compatible Responses API now accepts a reasoning effort setting on non-thinking models, and llama-server is left to infer the draft model's speculative decoding type on its own.
- Ollama v0.40.6 adds the q2_0 GGUF tensor type and fixes orphaned runner processes on Windows
Ollama published v0.40.6 on GitHub on October 16. The release adds the q2_0 tensor and file types to its GGUF support, prevents orphaned runner processes on Windows, and switches --ask mode to identifiers for question keys.
- Ollama v0.40.5-rc1 ports MLX Metal custom kernels to CUDA and accepts custom tool calls in Responses history
Ollama's v0.40.5-rc1 pre-release ports MLX Metal custom kernels to CUDA, bumps MLX, clarifies cloud payment errors in the app, standardizes model recommendation docs, and lets the OpenAI-compatible Responses path accept custom tool calls in history.
- Ollama v0.40.4-rc0 updates llama.cpp for decision models and EmbeddingGemma 2 and lists installed models in Codex's model picker
The v0.40.4-rc0 pre-release updates the bundled llama.cpp to b11521 for decision models and EmbeddingGemma 2. The server now reuses verified manifests and loaded models in inference routes, and the MLX backend compacts restored recurrent state after speculative rollback. Ollama's launch command also includes installed models in the Codex /model picker.
- Ollama 0.40.1 proxies cloud usage and balance APIs and fixes Windows reads past 2GiB
Ollama's v0.40.1 release lets the server proxy cloud usage and balance APIs, fixes reads past 2GiB on Windows, removes the account step from CLI onboarding and avoids symlinks in manifests on Windows. It also drops a Metal residency patch for MLX now that the fix is upstream, per the GitHub release notes.