Agens Volundr 32B Preview
Agens Volundr 32B Preview is an open-weights model for reasoning, coding, and agent work, with support for text, images, tool calls, and English, Chinese, and Cantonese. The preview checkpoint is available under Apache-2.0, though Blockway identifies long agentic sessions as a weakness.
What it does
Developers can download and run it for coding, reasoning, and tool-using tasks, including tasks that involve images.
- Availability
- Preview checkpoint; full v1 release announced for 2026-10-18
- Pricing model
- Not established
- Deployment
- Runs on Blockway's sglang build
- Privacy
- Not established
- Licence
- Apache-2.0
Capability
Modality
Access
Deployment
Pricing
Target user
Evidence
6 claims- The Preview checkpoint is available from the Blockway/Agens-Volundr-32B-Preview repository.
- The model handles text and images and calls tools. [1]
Volundr speaks English, Chinese (Simplified and Traditional) and Cantonese; handles text and images; calls tools; and ships with Blockway's control tokens and chat template (see Format and serving).
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model is built for reasoning, coding, and agent work. [1]
**Agens Volundr 32B** is an open-weights model from **Blockway**, built for reasoning, coding and agent work.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The model runs on Blockway's sglang build. [1]
**Serving.** Volundr runs on the Blockway sglang build (model class, `agens` parsers, reasoning budget).
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The full Agens Volundr 32B v1 release is announced as out on 2026-10-18. [1]
> **Agens Volundr 32B v1 is out (2026-10-18):** the full release this Preview announced — agent post-training, 0/50 repetition loops, session continuation, INT4 / GGUF / drafter and a 16 GB Engine build.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights. - The Preview fell into repetition loops in 36 of 50 SWE-bench tasks in Blockway's runs. [1]
- Long agentic coding sessions: in our SWE-bench runs through Codeway, Volundr fell into repetition loops in 36 of 50 tasks (output repeating a line, or tool calls written as plain text), against none for Qwen3.8-27B and Agens Pilot.
Hugging Face. Quoted text is outside CC BY 4.0 and keeps its original rights.
Sources
1
1
Hugging FaceOwner-controlled
https://huggingface.co/Blockway/Agens-Volundr-32B-Preview
Retrieved 2026-10-28 00:51 UTC
Verified
Versions
4| Version | Effective | Change | Changed fields | State |
|---|---|---|---|---|
| 4 | 2026-10-19 08:20 UTC | evidence update | score | Current |
| 3 | 2026-10-18 09:19 UTC | material change | capability, modality, access, availability, deployment | Current |
| 2 | 2026-10-06 20:09 UTC | material change | modality, access, availability, deployment | Current |
| 1 | 2026-10-05 21:08 UTC | publication | None | Current |
In briefs
1- Build and govern workplace agents with North 2 Daily brief 2026-10-05
Original catalog content is licensed under CC BY 4.0, attributed to wowsuchai. Source quotations and linked content keep their original rights.