# Swift 1.5 Qwen3.8-27B (slim)

> Swift 1.5 Qwen3.8-27B is UkisAI's post-trained Qwen3.8-27B that aims to think in fewer tokens, a 27B dense vision-language model with the base's 262,144-token context, published in BF16 with GGUF, AWQ, NVFP4, FP8 and MLX builds under UkisAI's own licence.

- Full: https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b.md (~10,450 tokens) · this version ~2,180 tokens · JSON https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b.json · canonical https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b#pricing · reviews https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #49 of 100 open-weight models · #42 in Open-weight models · not agent-ready · confidence low**


Assessment: A Qwen3.8-27B cousin post-trained to think in fewer tokens, for coding and agent work with images in, that fits a 24 GB machine with UkisAI's own 4-bit builds. Commercial use is free only below US$1M revenue. The main risks are that nobody at UkisAI answers its discussions, and that the shorter thinking rests on UkisAI's own benchmarks.

## Facts

- Kind: Open-weight model · vendor: UkisAI · category: Open-weight models · legal entity: not named · provenance 13/100
- Packages: huggingface `ukisai/Swift-1.5-Qwen3.8-27b`, huggingface `ukisai/Swift-1.5-Qwen3.8-27B-GGUF`
- Auth: None · pricing: Free · x402: no · licence: Swift Open License v1.0 (custom, not an OSI licence) on UkisAI's changes, free for commercial use below US$1M gross annual revenue with a separate Swift Enterprise License above it. Built on Qwen3.8-27B, whose Apache 2.0 still governs the base
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/ukisai/Swift-1.5-Qwen3.8-27b and UkisAI's quantised repositories, read from the model card, the files and their metadata, the LICENSE, NOTICE and bundled Apache licence, UkisAI's product page and the runtimes' own code on 21 October 2026. Nothing was downloaded or run. UkisAI's research API isn't graded
- Built on: Qwen/Qwen3.8-27B (Apache 2.0), declared in the card's metadata as a fine-tune, by way of Swift 1.0. Same architecture and config, new weights in every shard
- Model: 27.8B parameters in the files (vision encoder and the base's multi-token prediction head included), 64 layers in Qwen3.8's mix of Gated DeltaNet and Gated Attention. Text, images and video in, text out
- Files: 18 safetensors shards in BF16 (55.6 GB) with chat_template.jinja, configs and tokenizer from the base, LICENSE, LICENSE-APACHE-2.0, NOTICE and a benchmark record. No pickle files and no Python files. Hugging Face lists a SHA-256 for every LFS file. The repository's history was squashed to an initial release on 24 September and a README update on 3 October
- Licence: Swift Open License v1.0, Apache-style, with commercial use licensed only below US$1,000,000 gross annual revenue (non-profits exempt for non-commercial use) and automatic termination on breach. It applies to UkisAI's contribution only, and says nothing in it limits rights in Qwen3.8-27B under Apache 2.0, a copy of which ships in the repository
- Runtimes: The card gives commands for Transformers, vLLM and SGLang with the qwen3_coder tool parser, and GGUF builds for llama.cpp. The architecture is qwen3_5, which released llama.cpp, Ollama, vLLM, SGLang and Transformers run. No Ollama or LM Studio library entry for Swift
- Context: 262,144 tokens native, per config.json and the card's serving commands, the same as the base
- Tools and reasoning: The base's chat template, which declares tools in `<tools>` and renders several `<tool_call>` blocks per turn, read by the qwen3_coder parser named in the card's commands. Thinking on by default with Qwen3.8's reasoning_effort (xhigh, medium, low), which the card benchmarks at each level
- Hardware: The card states no memory needs and serves BF16 on one GPU with tensor parallel 1. The publisher's 4-bit builds are 15.9 GB (MLX, patched loader), 16.4 GB (GGUF IQ4_XS with projector), 18.4 GB (GGUF Q4_K_M with projector) and about 19.5 GB (AWQ and AutoRound)
- Training data: A general description. RL and on-policy distillation that penalise tokens linked to overthinking, seeded from ukisai/Qwen3.8-27B-multi-turn-agent-sft (about 15,200 Terminus-2 agent traces from Qwen3.8-27B on OpenThoughts-Agent tasks), which the card says was re-sampled into RL environments rather than used as is. No cut-off

- Scores: Capability pending, Reach 80, Risk 50, Agent interface 60, Documentation and upkeep 43 · total over the 7 assessed categories
- Why:  · Reach, Memory 25 of 30, UkisAI's 4-bit MLX build is 15.9 GB and its 4-bit GGUF with the vision projector 16.4 GB, both under 20 GB with a fifth add… · Risk, File safety 25, safetensors only and nothing from the repository runs. · Agent interface, Chat template 20, the base's Jinja file. · Documentation and upkeep, Model card 8 of 20, intended use, parameters and context only.
- Sources: 16, open questions: 4, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/swift-1-5-qwen3-8-27b.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/swift-1-5-qwen3-8-27b.svg` or a link to https://www.anchorterminal.com/tools/swift-1-5-qwen3-8-27b from a page on ukisai.com or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Quickest start is `llama-server -hf ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Q4_K_M --jinja`, about 18.4 GB with the vision projector
2. On our reference machine BF16 fits, `vllm serve ukisai/Swift-1.5-Qwen3.8-27b --max-model-len 262144 --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder`
3. The template is Qwen3.8-27B's, so `chat_template_kwargs` with `reasoning_effort` low, medium or xhigh (default) and `enable_thinking` false work as on the base
4. UkisAI benchmarked it at temperature 1.0, top_p 0.95, top_k 20 and presence_penalty 0, which is a reasonable starting point
5. The MLX 4-bit build needs UkisAI's patched MLX-LM and has no image or video chat
6. Check the licence before commercial use, it stops at US$1M gross annual revenue including affiliates
7. MTP self-speculative decoding is available with the base's head, flags on the card

## Connect

```bash
vllm serve ukisai/Swift-1.5-Qwen3.8-27b --max-model-len 262144 --reasoning-parser qwen3 \
  --enable-auto-tool-choice --tool-call-parser qwen3_coder
# or the 4-bit GGUF with its vision projector:
llama-server -hf ukisai/Swift-1.5-Qwen3.8-27B-GGUF:Q4_K_M --jinja
```

```bash
curl http://localhost:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "ukisai/Swift-1.5-Qwen3.8-27b", "messages": [{"role": "user", "content": "Hello, Swift."}], "chat_template_kwargs": {"reasoning_effort": "low"}}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |


## Panel reviews (4, average 2.3/5, desk reviews from public material, no calls made)

- ★★☆☆☆ History squashed to two commits (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★☆☆ Free below US$1M revenue, then an unpriced licence (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Reasoning effort mentioned, never explained (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Clean safetensors under a revenue-capped licence (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings from 2 to 3. They agree UkisAI's clean safetensors sit under a custom licence that stops free commercial use at US$1,000,000 revenue, and that the card's metadata claims a gate the repository doesn't have. Read it as a reasoning-efficient Qwen3.8-27B with many UkisAI builds for a 24 GB card, no independent results and no replies.