GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
ayoubkirouane
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
ayoubkirouane
Model Tree
Qwen/Qwen3.6-35B-A3B
Input Modalities
Output Modalities
Supported Functionality
Trace-language steerability is partial: an explicit instruction to reason in English on a Greek question is obeyed on 44.8% of items (the reverse, 83.7%). A deployment that must force the trace language should test this axis, or use Sophea-Qwen3.6-v1.1 (53.9%). Not evaluated for safety-critical, legal, or medical use.
Fine-tuned from Qwen3.6-35B-A3B; shipped merged to dense weights.
Arrows mark the desirable direction for every metric: ↑ higher is better, ↓ lower is better.
The pooled accuracy delta (−6.9) is a single-run reading inside the 7.7 pp seed-to-seed noise floor measured on this family; the paper discloses it and does not interpret it. Difficulty adaptation: the base overruns (≥3× the per-item median trace length) on 98.0% of easy items and its rate is identical on easy and hard items; the fine-tunes overrun on 0.2–12% of easy items and spend more only when the item is harder.
No fine-tune in the study beats its own base on any axis of this table. Conditioning on rows that emit the requested answer line, the release reads 95.8 / 78.8 / 39.5 against the base's 96.0 / 82.8 / 42.7 (math / commonsense / logic), so a substantial share of the raw gap is format compliance, not reasoning.
This suite exists to answer one question: did the Greek fine-tune damage general ability? It did not. Against its own base, across 14 bilingual NLU benchmarks: −0.01 Greek / +0.08 English, statistically flat. Read the numbers below as a forgetting check, not a capability leaderboard; movements of a point or two are within per-benchmark noise on this suite.
Register control and grammaticality, judge-scored at n=46/58 (directional evidence only): register 45 → 42 of 46, grammaticality 41 → 42 of 58; neither politeness control nor grammatical Greek is lost.
Serve with vLLM (OpenAI-compatible; the weights are dense, no adapter needed):
bash
vllm serve KIEFERSA/Sophea-Qwen3.6-v1 --served-model-name sophea-qwen3.6-v1 --reasoning-parser qwen3
Client (OpenAI SDK), thinking left ON:
python
from openai import OpenAIclient = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")resp = client.chat.completions.create(model="sophea-qwen3.6-v1",messages=[{"role": "user", "content": "Ένα κατάστημα πουλά στυλό προς 1,50 € το ένα. Πόσο κοστίζουν 12 στυλό;"}],temperature=0.7,extra_body={"chat_template_kwargs": {"enable_thinking": True}}, # thinking model)print(resp.choices[0].message.reasoning_content) # the trace, in the question's languageprint(resp.choices[0].message.content) # the final answer
Transformers:
python
import torchfrom transformers import AutoModelForCausalLM, AutoTokenizertok = AutoTokenizer.from_pretrained("KIEFERSA/Sophea-Qwen3.6-v1")model = AutoModelForCausalLM.from_pretrained("KIEFERSA/Sophea-Qwen3.6-v1", torch_dtype=torch.bfloat16, device_map="auto")messages = [{"role": "user", "content": "Ένα τρένο διανύει 240 χλμ σε 3 ώρες. Ποια είναι η μέση ταχύτητά του;"}]text = tok.apply_chat_template(messages, tokenize=False,add_generation_prompt=True, enable_thinking=True)inputs = tok(text, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=2048, temperature=0.7, do_sample=True)print(tok.decode(out[0, inputs.input_ids.shape[1]:], skip_special_tokens=False))
Steering the trace language (partial, measured per checkpoint): append "Think step by step IN ENGLISH." to a Greek question, or "Σκέψου βήμα προς βήμα ΣΤΑ ΕΛΛΗΝΙΚΑ." to an English one. This checkpoint follows the English steering suffix on 44.8% of Greek items and the Greek suffix on 83.7% of English items.
This model ships the multi-token-prediction head: 19 mtp.* tensors (~1.7 GB, bf16) in
model-mtp.safetensors, the single-layer draft stack that config.json declares through
mtp_num_hidden_layers: 1. The head is the base model's own; speculative decoding verifies every
drafted token against the main model, so it changes throughput only, never the output
distribution.
Enable it with vLLM:
bash
vllm serve KIEFERSA/Sophea-Qwen3.6-v1 --served-model-name sophea-qwen3.6-v1 --trust-remote-code \--speculative-config '{"method":"qwen3_5_mtp","num_speculative_tokens":2}'
The full vision stack of the multimodal base is included, byte-identical to the base: image inputs work exactly as on Qwen3.6-35B-A3B. The adaptation touched only the language decoder; vision performance has not been separately re-evaluated for this release.
Apache-2.0. The base model, Qwen/Qwen3.6-35B-A3B, is released under Apache-2.0 and this fine-tune inherits those terms. Verify base-model terms before use.
This model is a release artifact of arXiv:2608.17744. If you use it, please cite:
bibtex
@misc{kirouane2026thinking,title={Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See},author={Ayoub Kirouane and Christos Petrocheilos},year={2026},eprint={2608.17744},archivePrefix={arXiv},primaryClass={cs.CL},url={https://arxiv.org/abs/2608.17744},}