Ollama – v0.17.6

🚨 Ollama v0.17.6 is out — and it’s a quick but important patch! 🚨

This release is light on features, heavy on precision:

🔧 Bug fix: Corrected how `glm-ocr` image tags are parsed in renderer prompts

🔗 PR #14584 by @Victor-Quqi

✅ Why it matters:

• If you’re using GLM-OCR (especially for multimodal OCR tasks), image tags like `<image>` in your prompts will now render correctly instead of causing errors or misinterpretations.
• Ensures smoother integration in custom renderer workflows — critical for anyone building multimodal apps or pipelines on top of Ollama.

📦 No new models, no API changes — just a clean, targeted fix to keep your local LLM workflows humming.

If you rely on GLM-OCR or custom multimodal prompts, update away! 🛠️

Let me know if you want a breakdown of how Ollama renderers work or how to test this fix! 🤖✨

🔗 View Release

Voxtral Wyoming – v1.0.0

🚨 Voxtral Wyoming v1.0.0 is live — and it’s production-ready! 🚀

The wait is over: this release marks the stable, final v1.0.0 of Voxtral Wyoming — your go-to offline STT service powered by Mistral’s Voxtral models, now fully integrated with Home Assistant Assist via the Wyoming protocol.

✨ What’s new (and why it matters):

✅ Stable & battle-tested — all major bugs squashed, performance optimized for real-world use

✅ API finalized — no more breaking changes ahead; integrations are safe to lock in

✅ Full tooling in place — docs, tests, and CI/CD pipelines are now rock-solid

✅ Zero flash, all function — no flashy new features, just a polished, reliable upgrade ready for production 🛠️

🎯 Whether you’re running it on CPU, CUDA (NVIDIA), or MPS (Apple Silicon), and whether your audio comes in MP3, OGG, FLAC, or WAV — Voxtral Wyoming handles it all with automatic PCM16 conversion. Config via env vars? Yep — host, port, language, model ID… all covered.

📦 Dockerized. Deployed. Ready.

🟢 Green light for production! Let’s build smarter, offline-first voice assistants — together. 🎤💡

🔗 View Release

Ollama – v0.17.5

🚨 Ollama v0.17.5 is live! 🚨

Hey AI tinkerers — fresh update alert! 🔥 Ollama just rolled out v0.17.5, and it’s a quiet but mighty one — especially if you love playing with Qwen3 or importing GGUF models. Here’s the lowdown:

🔹 GGUF love, expanded! 🎁

• Full support for importing and running Qwen3 models (like `Qwen3-0.6B`, `Qwen3-1.7B`) — straight from Hugging Face or wherever you grab your GGUFs.
• Smoother imports, fewer hiccups 🛠️

🔹 Under-the-hood polish

• Bug fixes and stability tweaks (you won’t see them, but you’ll feel the smoother run).

💡 Why care?

If you’re experimenting with lightweight Qwen3 variants or love the flexibility of GGUF (quantized, portable, efficient 📦), this update makes your workflow just a little more magical. ✨

Ready to upgrade? `ollama pull ollama` 🚀

Let us know how it runs!

🔗 View Release

Voxtral Wyoming – v0.5.0

_New update detected._

🔗 View Release

Voxtral Wyoming – v0.4.0

_New update detected._

🔗 View Release

Lemonade – v9.4.1

_New update detected._

🔗 View Release

Voxtral Wyoming – v0.3.0

_New update detected._

🔗 View Release

Ollama – v0.17.4

🚀 Ollama v0.17.4 is live! Here’s what’s fresh in this patch release:

🔹 Stable Tool Calling for GLM-4 & Qwen3

✅ Reliable tool/function calling support—no more misaligned or garbled tool outputs!

✅ Works seamlessly with `curl`, Python clients, and custom tools via the Ollama API.

🔹 Better JSON & Parser Handling

🧠 Internal upgrades to model parsers—especially for Chinese-language models (GLM, Qwen).

📊 More consistent parsing of JSON-formatted tool responses.

🔹 Minor Fixes & Tweaks

⚙️ Performance bumps, bug fixes, and general polish—zero breaking changes.

Perfect for anyone relying on structured outputs or tool integrations with local LLMs. Try it out and let us know how your tool-calling workflows feel! 🛠️✨

🔗 View Release

Ollama – v0.17.3: model: fix qwen3 tool calling in thinking (#14477)

🚨 Ollama v0.17.3 is live — and it’s fixing a big one for Qwen3 fans! 🎯

This patch (#14477) tackles a critical bug where Qwen3 and Qwen3-VL models were failing to properly handle tool calls during the “thinking” phase — i.e., before “ closes.

🔧 What’s fixed?

✅ Tool-call detection now works mid-think: The model correctly spots `<tool_call>` (tool call start tag) while still in thinking mode and smoothly transitions into tool-parsing — matching Hugging Face Transformers behavior.

✅ Robust tag parsing: Handles overlapping or partial tags (e.g., `<tool_call>` appearing before “) without breaking.

✅ Streaming-safe: Works reliably even when `<tool_call>` is split across chunks in streaming responses.

🧠 Why you’ll care:

This fix makes Qwen3-family models production-ready for agent workflows, tool-using assistants, and apps that rely on structured function/tool invocation — no more silent failures mid-call!

📦 Update now:

“`bash

ollama pull qwen3 # for text models

ollama pull qwen3vl # for vision-language variants

“`

Happy tool-calling! 🛠️✨

🔗 View Release

Ollama – v0.17.2

🚨 Ollama v0.17.2 is live! 🚨

Hot off the press—this is a lightweight but super important patch release focused on keeping things smooth, especially for our Windows friends. 💻✨

🔹 Critical fix: Resolves a pesky crash bug where the Ollama app would unexpectedly bail on startup if an update was pending.

✅ Now, updates flow seamlessly—no more “why won’t it open?!” moments.

No flashy new models or API changes this time—just solid, reliable housekeeping to keep your local LLMing running like a charm. 🛠️✨

Upgrade soon and say goodbye to launch-day surprises! 🎉

Let me know if you want a quick refresher on how to update or try out the latest models. 🚀

🔗 View Release