Ollama app icon

Ollama Preview Build

Download the latest notarized macOS preview for Gemma 4, Apertus 1.5 Omni, GPT-OSS, and MLX media work, or build the source preview locally.

These preview builds target Apple Silicon Macs and exercise MLX-backed model runners before the changes are part of a regular Ollama release.

Automatic update checks and startup update behavior are disabled for the preview build; manual update actions remain user-initiated.

Ollama 0.34.1 MLX preview

Published September 18, 2026 · signed and notarized macOS DMG
Latest
File
Ollama-0.34.1-30-gc8b964e-mlx-preview-v0.34.1.dmg
Size
192.5 MB
SHA-256
bf0da003659be88682e1605628364d559c2cddffabae6f52f18cde57ae94f845
Upstream
Ollama v0.34.1 @ 38fdb5dd58c7
Highlights
Official Ollama v0.34.1Updated official baseline with the current app, API, model-create, and runtime behavior from the stable v0.34.1 release. Included MLX preview capabilitiesNative GPT-OSS and Apertus 1.0/1.1 Mini runtimes with validated thinking and tool calling
Gemma 4 and Apertus 1.5 Omni image/audio with MP3, WebP, cancellation, and bounded media processing
Responses namespace tools, custom apply_patch compatibility, sparse tokenizer allocation, and desktop audio attachments. Release qualificationThe exact signed, notarized, and stapled DMG passed capability, compatibility, balanced performance, memory, cleanup, and Gatekeeper checks. Existing public model aliases remain unchanged.
Download DMG

Ollama 0.33.3 MLX preview

Published September 5, 2026 · signed and notarized macOS DMG
Previous
File
Ollama-0.33.3-75-g9581bea-mlx-preview-v0.33.3.dmg
Size
155.9 MB
SHA-256
529fd4c2fc81109e53b8e94db88a74ff6d339f10b1758d748bcaa7c7ab160063
Upstream
Ollama v0.33.3 @ b79067b0db74
Highlights
Official Ollama v0.33.3New since v0.33.2 preview: Gemma 4 image/audio support on MLX, cached prompt-token reporting, model-defined GGUF defaults, and updated MLX, MLX-C, and llama.cpp. Included MLX preview capabilitiesNative GPT-OSS and Apertus runtimes, including retained Apertus Mini 4B behavior
Gemma 4 and Apertus 1.5 Omni image/audio with MP3, WebP, and bounded media handling
Thinking, tool calling, Responses namespaces, custom apply_patch, and desktop audio attachments across validated preview models.
Download DMG

Ollama 0.33.2 MLX preview

Published August 31, 2026 · signed and notarized macOS DMG
Previous
File
Ollama-0.33.2-90-g6c08853-mlx-preview-v0.33.2.dmg
Size
155.9 MB
SHA-256
37fbe56a247acb85cadab01422e7f625544195f056a8dfff7a31f9555bc83422
Upstream
Ollama v0.33.2 @ f96e7aa0513b
Highlights
Official Ollama v0.33.2New since v0.33.1 preview: updated official baseline with desktop and Claude-proxy behavior refinements plus discovery and ML API cleanup. Included MLX preview capabilitiesNative GPT-OSS and Apertus runtimes, including retained Apertus Mini 4B behavior
Gemma 4 and Apertus 1.5 Omni ordered image/audio with bounded media handling
Thinking, tool calling, Responses namespaces, custom apply_patch, and desktop audio attachments across validated preview models. Release processThis first refresh used two collaborating AI agents to reconstruct and rebase the PR stack step by step, then validate functionality mostly autonomously. The user retained approval, signing, and publication decisions. Performance evidence is thermally inconclusive, so no quantitative performance claim is made.
Download DMG

Ollama 0.33.1 MLX preview

Published August 28, 2026 · signed and notarized macOS DMG
Previous
File
Ollama-0.33.1-96-g6ffd70f-mlx-preview-v0.33.1.dmg
Size
155.9 MB
SHA-256
c18edd7ddff69027dac052b396d83bc2715c90c32985743113206cdb963c7841
Upstream
Ollama v0.33.1 @ 13f2fb8c9927
Highlights
Official Ollama v0.33.1 New since v0.32.15 preview: Claude Desktop integration, more reliable resumed-prefill caching, structured MLX output, Qwen3.8 Flash Next, updated MLX and llama.cpp runtimes, and safer loading from slow storage. Included MLX preview capabilities Native GPT-OSS and Apertus runtimes, including Apertus 1.1 Mini affine imports
Gemma 4 and Apertus 1.5 Omni ordered image/audio with quantized media, MP3, WebP, cancellation, and bounded processing
Thinking, tool calling, Responses namespaces, and custom apply_patch support across validated preview models.
Download DMG

Ollama 0.32.15 MLX preview — r2

Published August 23, 2026 · signed and notarized macOS DMG
Previous
File
Ollama-0.32.15-78-g5b266ac-mlx-preview-v0.32.15-r2.dmg
Size
150.1 MB
SHA-256
6ed699b80b8b2babe0082ee76c00a5fcb02926fdadac182858e8f5c0f7c3b809
Upstream
Ollama v0.32.15 @ b7871fc0d1d8
Highlights
v0.32.15 r2 with native GPT-OSS MXFP4 preservation, packed Apertus 70B MXFP8, media models, and Responses tools. Official Ollama v0.32.15 New since v0.32.13 preview: desktop onboarding, safer mid-stream parser failures, WebP transcoding for llama-server, more tolerant Qwen system-message placement, and cached model metadata. Included MLX preview capabilities Native GPT-OSS and Apertus runtimes, including Apertus 1.1 Mini imports and published artifacts under conversation-quality investigation
Gemma 4 and Apertus 1.5 Omni image/audio with MP3, WebP, bounded media handling, and desktop audio attachments
Thinking, tool calling, and Responses namespace/custom apply_patch support across validated preview models. r2 release fixes Preserves native GPT-OSS MXFP4 snapshots and imports Apertus 70B as packed MXFP8. The signed DMG freshly created and ran GPT-OSS 20B/120B plus Apertus 70B.
Download DMG

Ollama 0.32.13 MLX preview — Codex apply_patch update

Published August 22, 2026 · signed and notarized macOS DMG
Previous
File
Ollama-0.32.13-74-gcb9b2ae-mlx-preview-v0.32.13-applypatch1.dmg
Size
150.2 MB
SHA-256
304a403bb320e22e9707aaf086df68daebfa57477a5501aea440b672644b2952
Upstream
Ollama v0.32.13 @ 0f25c31bd53b
Highlights
v0.32.13 base with desktop audio attachments, Responses apply_patch compatibility, and the complete MLX model stack. Official Ollama v0.32.13 Qwen 3.8 27B with Apple Silicon MLX optimization and developer-instruction support. Included MLX preview capabilities Native GPT-OSS and Apertus runtimes, including Apertus 1.1 Mini INT2–INT8 imports
Gemma 4 and Apertus 1.5 Omni image/audio with MP3, WebP, and bounded media handling
Thinking and tool calling across the supported preview models. Desktop audio attachments Select or drop MP3 and WAV files in chat; supported audio models receive the media bytes through the established chat transport. Codex apply_patch compatibility Exposes Codex custom apply_patch tools through the Responses API, while keeping ordinary non-GPT-OSS models on their supported tool path.
Download DMG

Older Preview Builds

Archived preview releases since May 1, 2026, starting with Ollama 0.22.1
Open Archive

To verify a download, run shasum -a 256 <downloaded-file.dmg> and compare it with the SHA-256 value shown above.

AI assistance disclosure: this page's release notes and explanatory copy were edited with AI assistance and reviewed before publication.