Sophea-Qwen3.6-v1

<div style="display:flex;align-items:center;gap:18px;padding:16px 20px;margin-bottom:16px;border-radius:14px;background:linear-gradient(100deg,#e8f5ec 0%,#f7fbf8 62%);border:1px solid #c6e7d1"><img alt="KIEFERSA" style="flex:0 0 auto;height:58px;width:58px;border-radius:12px;object-fit:contain" src="data:image/webp;base64,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"/><div><div style="font-size:1.24rem;font-weight:700;color:#003713;line-height:1.2">Sophea-Qwen3.6-v1</div><div style="font-size:.86rem;color:#3a5a47;margin-top:2px">Greek and English reasoning fine-tune of Qwen3.6-35B-A3B: thinking MoE, vision retained, MTP speculative decoding</div></div></div>

📄 Released with the paper Thinking in a Low-Resource Language: What SFT Builds, What RL Fixes, What Accuracy Cannot See (arXiv:2608.17744). Every number on this card is measured, contextualized, and caveated in the paper; read it for the full evaluation protocol.

• Creator: Kiefer SA (Sophea AI Lab, Athens)
• Base model: Qwen3.6-35B-A3B (sparse MoE transformer, 36.0B total / 3.97B active, 8-of-256 routing)
• Ships with: the base's full multimodal (vision) stack, untouched, and the multi-token-prediction draft head for speculative decoding
• Languages: Greek and English (the reasoning trace follows the question's language)
• Decoding: thinking (enable_thinking=true); the trace is the product and follows the question's language

Serve thinking. Leave enable_thinking=true so the reasoning trace is produced; the trace follows the question's language. Over an OpenAI-compatible vLLM endpoint, add --reasoning-parser qwen3 so the trace arrives in message.reasoning_content and the answer in message.content.

Intended use

• Greek and English reasoning assistant: math, science, logic, and general step-by-step problem solving
• Deployments that need the reasoning trace itself (auditable working, tutoring, verification), in the user's language

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.

---

Evaluation

Arrows mark the desirable direction for every metric: ↑ higher is better, ↓ lower is better.

Headline scorecard

<div style="overflow-x:auto"> <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif"> <thead><tr> <th style="padding:10px 8px;text-align:left;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Axis</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Qwen3.6-35B-A3B (base)</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Sophea-Qwen3.6-v1</th> </tr></thead><tbody> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Greek-trace fidelity, Greek questions (ratio &ge; 0.9, n=5,156) ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.0% (0/5,156)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">97.98% (5,052/5,156)</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">English traces on English questions ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">100%</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">100% (n=1,900)</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">In-question language switches / 100 sentences ↓</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">16.4</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">0.0</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Answer-channel leak ↓</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">15.6% (English control)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">3.5%</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Answer-format fallback, Greek think lane ↓</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.2&ndash;3.3% per domain</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">24.1%</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Instructed override, el&rarr;EN / en&rarr;EL ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">n/a (already reasons in English)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">44.8% / 83.7%</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Median trace length, tokens (own tokenizer) ↓</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">1,788</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">586 (3.0&times; fewer)</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Greek NLU macro, forgetting check (vs own base) ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">75.7</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">75.6 (&minus;0.01 unrounded)</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">English NLU macro, forgetting check (vs own base) ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">85.1</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">85.2 (+0.08 unrounded)</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Pooled Greek reasoning accuracy (n=5,156) ↑</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">80.6</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">73.7</td></tr> </tbody></table></div>

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.

Per-benchmark detail

Per-domain, Greek think lane

<div style="overflow-x:auto"> <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif"> <thead><tr> <th style="padding:10px 8px;text-align:left;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Domain (Greek think lane)</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Base acc ↑</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Sophea-Qwen3.6-v1 acc ↑</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Base fallback ↓</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Sophea-Qwen3.6-v1 fallback ↓</th> </tr></thead><tbody> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Math (two math slices)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">92.9</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">83.3</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">3.3%</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">18.2%</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Commonsense (HellaSwag + Winogrande)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">82.4</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">75.8</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.5%</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">22.9%</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Logic (ProofWriter, macro-recall)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">42.6</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">37.3</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.2%</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">46.3%</td></tr> </tbody></table></div>

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.

Forgetting check: Titan-1 suite (non-reasoning mode)

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.

<div style="overflow-x:auto"> <table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif"> <thead><tr> <th style="padding:10px 8px;text-align:left;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Benchmark (Titan-1 suite)</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Base</th> <th style="padding:10px 8px;text-align:center;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;">Sophea-Qwen3.6-v1</th> </tr></thead><tbody> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">ARC-Challenge-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">91.8</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">92.0</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">ARC-Easy-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">96.9</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">97.1</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Belebele-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">92.7</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">93.7</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">GreekMMLU</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">83.8</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">83.0</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">HellaSwag-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">57.6</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">59.4</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Medical-MCQA-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">81.2</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">80.3</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">TruthfulQA-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">39.1</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">36.5</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Winogrande-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">61.0</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">62.5</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">MMLU-el</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">76.9</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">76.2</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);"><em>Greek macro</em></td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);"><em>75.7</em></td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;"><em>75.6</em></td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">ARC-Challenge (en)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">96.1</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">95.7</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">ARC-Easy (en)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">99.1</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">99.1</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">HellaSwag (en)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">73.2</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">74.2</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Winogrande (en)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">73.2</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">74.2</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">MMLU (en)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">83.9</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;">82.6</td></tr> <tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);"><em>English macro</em></td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);"><em>85.1</em></td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;"><em>85.2</em></td></tr> </tbody></table></div>

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.

Usage

Serve with vLLM (OpenAI-compatible; the weights are dense, no adapter needed):

vllm serve KIEFERSA/Sophea-Qwen3.6-v1 --served-model-name sophea-qwen3.6-v1 --reasoning-parser qwen3

Client (OpenAI SDK), thinking left ON:

from openai import OpenAI

client = 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 language
print(resp.choices[0].message.content)            # the final answer

Transformers:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tok = 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.

Speculative decoding (MTP)

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:

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}'

Vision

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.

License

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.

Citation

This model is a release artifact of arXiv:2608.17744. If you use it, please cite:

@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},
}