Ollama v0.40.0-rc6

Use Ollama to run supported models locally on Apple Silicon, including decision and embedding models supported by MLX.

Availability

Released as v0.40.0

Pricing model

Not established

Deployment

On Apple Silicon devices

Privacy

Not established

Licence

Not established

• The announcement identifies the release as v0.40.0-rc6. ^[1]
• The release implements the EmbeddingGemma2Model architecture on the MLX runner. ^[1]

  Implements the EmbeddingGemma2Model architecture on the MLX runner: 24-layer

• The architecture uses a 24-layer bidirectional text encoder with PLE. ^[1]

  bidirectional text encoder with PLE, shared gemma4 vision/audio towers,

• The architecture uses shared Gemma 4 vision and audio towers. ^[1]
• The output uses mean pooling and L2. ^[1]
• The /api/embed endpoint accepts per-item media via input dictionaries. ^[1]

  /api/embed accepts per-item media via input dicts.

• The release includes gemma4, qwen3.6, and qwen3.5. ^[2]

  Additional models include gemma4 , qwen3.6 and qwen3.5

• On Apple Silicon, architectures supported by the MLX runtime automatically run on MLX. ^[2]

  In this release, on Apple Silicon devices, model architectures supported by the MLX runtime will automatically run on MLX.

• Decision models Nimble, tev1, clef, and clef-flash are available on MLX. ^[2]

  Decision models are now available on MLX as well: Nimble tev1 clef clef-flash

• MLX supports the embedding model embeddinggemma-2. ^[2]

  In this release, on Apple Silicon devices, model architectures supported by the MLX runtime will automatically run on MLX.

• Ollama v0.40.0 was released. ^[2]

• Edit generated images through conversation with Gemini Nano Banana 2.1 Daily brief 2026-10-06