Qwen3.5-4B

MERA Created at 12.05.2026 00:03
0.567
The overall result
94
Place in the rating
Weak tasks:
582
RWSD
420
PARus
234
RCB
341
ruEthics
500
MultiQ
153
ruWorldTree
204
ruOpenBookQA
303
CheGeKa
111
ruMMLU
426
ruHateSpeech
541
ruDetox
255
ruHHH
257
ruTiE
122
ruHumanEval
205
USE
29
MathLogicQA
28
ruMultiAr
320
SimpleAr
65
LCS
340
BPS
30
ruModAr
108
MaMuRAMu
114
ruCodeEval
+19
Hide

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Metric
LCS 0.172 Accuracy
RCB 0.534 / 0.428 Accuracy F1 macro
USE 0.176 Grade norm
RWSD 0.4 Accuracy
PARus 0.614 Accuracy
ruTiE 0.672 Accuracy
MultiQ 0.188 / 0.158 F1 Exact match
CheGeKa 0.083 / 0.077 F1 Exact match
ruModAr 0.958 Exact match
MaMuRAMu 0.785 Accuracy
ruMultiAr 0.938 Exact match
ruCodeEval 0.18 / 0.513 / 0.659 Pass@k
MathLogicQA 0.969 Accuracy
ruWorldTree 0.945 / 0.775 Accuracy F1 macro
ruOpenBookQA 0.83 / 0.71 Accuracy F1 macro

Evaluation on open tasks:

Go to the ratings by subcategory

The table will scroll to the left

Task name Result Metric
BPS 0.873 Accuracy
ruMMLU 0.702 Accuracy
SimpleAr 0.926 Exact match
ruHumanEval 0.176 / 0.492 / 0.652 Pass@k
ruHHH 0.669
ruHateSpeech 0.555
ruDetox 0.058
ruEthics
Correct God Ethical
Virtue 0.22 0.298 0.283
Law 0.229 0.289 0.278
Moral 0.231 0.316 0.29
Justice 0.194 0.285 0.265
Utilitarianism 0.154 0.249 0.232

Information about the submission

Mera version
v1.2.0
Torch Version
2.10.0
The version of the codebase
7c56310
CUDA version
12.8
Precision of the model weights
bfloat16
Seed
1234
Batch
1
Transformers version
4.57.6
The number of GPUs and their type
1 x NVIDIA A100-SXM4-80GB
Architecture
local-chat-completions

Team:

MERA

Name of the ML model:

Qwen3.5-4B

Model size

4.0B

Model type:

Opened

SFT

Additional links:

https://qwen.ai/blog?id=qwen3.5

Architecture description:

Type: Causal Language Model with Vision Encoder Training Stage: Pre-training & Post-training Language Model Number of Parameters: 4B Hidden Dimension: 2560 Token Embedding: 248320 (Padded) Number of Layers: 32 Hidden Layout: 8 × (3 × (Gated DeltaNet → FFN) → 1 × (Gated Attention → FFN)) Gated DeltaNet: Number of Linear Attention Heads: 32 for V and 16 for QK Head Dimension: 128 Gated Attention: Number of Attention Heads: 16 for Q and 4 for KV Head Dimension: 256 Rotary Position Embedding Dimension: 64 Feed Forward Network: Intermediate Dimension: 9216 LM Output: 248320 (Tied to token embedding) MTP: trained with multi-steps

Description of the training:

Unified Vision-Language Foundation: Early fusion training on multimodal tokens achieves cross-generational parity with Qwen3 and outperforms Qwen3-VL models across reasoning, coding, agents, and visual understanding benchmarks. Scalable RL Generalization: Reinforcement learning scaled across million-agent environments with progressively complex task distributions for robust real-world adaptability.

Pretrain data:

Trained on a significantly larger scale of visual-text tokens compared to Qwen3, with enriched Chinese/English, multilingual, STEM, and reasoning data under stricter filtering. This update prioritizes the expansion of low-resource languages. Through this broader linguistic scope, Qwen3.5 is dedicated to fostering global AI equity. Natively multimodal via early text-vision fusion and expanded visual/STEM/video data, outperforming Qwen3-VL at similar scales. Multilingual coverage grows from 119 to 201 languages/dialects; a 250k vocabulary (vs. 150k) boosts encoding/decoding efficiency by 10–60% across most languages.

License:

Apache license 2.0

Inference parameters

Generation Parameters:
rucodeeval - do_sample=true;temperature=0.6;until=[\"<|im_end|>\"],max_gen_toks=10000; ruhumaneval - do_sample=true;temperature=0.6;until=[\"<|im_end|>\"],max_gen_toks=10000; rutie_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; bps_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; chegeka - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; lcs_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; mamuramu_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; mathlogicqa_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; multiq - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; parus_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rcb_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rudetox - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; ruethics_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; ruhatespeech_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; ruhhh_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rummlu_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rumodar - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rumultiar - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; ruopenbookqa_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; ruworldtree_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; rwsd_gen - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; simplear - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000; use - do_sample=false,until=[\"<|im_end|>\"],max_gen_toks=10000;

The size of the context:
262144

System prompt:
Реши задачу по инструкции ниже. Не давай никаких объяснений и пояснений к своему ответу. Не пиши ничего лишнего. Пиши только то, что указано в инструкции. Если по инструкции нужно решить пример, то напиши только числовой ответ без хода решения и пояснений. Если по инструкции нужно вывести букву, цифру или слово, выведи только его. Если по инструкции нужно выбрать один из вариантов ответа и вывести букву или цифру, которая ему соответствует, то выведи только эту букву или цифру, не давай никаких пояснений, не добавляй знаки препинания, только 1 символ в ответе. Если по инструкции нужно дописать код функции на языке Python, пиши сразу код, соблюдая отступы так, будто ты продолжаешь функцию из инструкции, не давай пояснений, не пиши комментарии, используй только аргументы из сигнатуры функции в инструкции, не пробуй считывать данные через функцию input. Не извиняйся, не строй диалог. Выдавай только ответ и ничего больше.

Description of the template:
{%- set image_count = namespace(value=0) %} {%- set video_count = namespace(value=0) %} {%- macro render_content(content, do_vision_count, is_system_content=false) %} {%- if content is string %} {{- content }} {%- elif content is iterable and content is not mapping %} {%- for item in content %} {%- if 'image' in item or 'image_url' in item or item.type == 'image' %} {%- if is_system_content %} {{- raise_exception('System message cannot contain images.') }} {%- endif %} {%- if do_vision_count %} {%- set image_count.value = image_count.value + 1 %} {%- endif %} {%- if add_vision_id %} {{- 'Picture ' ~ image_count.value ~ ': ' }} {%- endif %} {{- '<|vision_start|><|image_pad|><|vision_end|>' }} {%- elif 'video' in item or item.type == 'video' %} {%- if is_system_content %} {{- raise_exception('System message cannot contain videos.') }} {%- endif %} {%- if do_vision_count %} {%- set video_count.value = video_count.value + 1 %} {%- endif %} {%- if add_vision_id %} {{- 'Video ' ~ video_count.value ~ ': ' }} {%- endif %} {{- '<|vision_start|><|video_pad|><|vision_end|>' }} {%- elif 'text' in item %} {{- item.text }} {%- else %} {{- raise_exception('Unexpected item type in content.') }} {%- endif %} {%- endfor %} {%- elif content is none or content is undefined %} {{- '' }} {%- else %} {{- raise_exception('Unexpected content type.') }} {%- endif %} {%- endmacro %} {%- if not messages %} {{- raise_exception('No messages provided.') }} {%- endif %} {%- if tools and tools is iterable and tools is not mapping %} {{- '<|im_start|>system\n' }} {{- "# Tools\n\nYou have access to the following functions:\n\n<tools>" }} {%- for tool in tools %} {{- "\n" }} {{- tool | tojson }} {%- endfor %} {{- "\n</tools>" }} {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }} {%- if messages[0].role == 'system' %} {%- set content = render_content(messages[0].content, false, true)|trim %} {%- if content %} {{- '\n\n' + content }} {%- endif %} {%- endif %} {{- '<|im_end|>\n' }} {%- else %} {%- if messages[0].role == 'system' %} {%- set content = render_content(messages[0].content, false, true)|trim %} {{- '<|im_start|>system\n' + content + '<|im_end|>\n' }} {%- endif %} {%- endif %} {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %} {%- for message in messages[::-1] %} {%- set index = (messages|length - 1) - loop.index0 %} {%- if ns.multi_step_tool and message.role == "user" %} {%- set content = render_content(message.content, false)|trim %} {%- if not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %} {%- set ns.multi_step_tool = false %} {%- set ns.last_query_index = index %} {%- endif %} {%- endif %} {%- endfor %} {%- if ns.multi_step_tool %} {{- raise_exception('No user query found in messages.') }} {%- endif %} {%- for message in messages %} {%- set content = render_content(message.content, true)|trim %} {%- if message.role == "system" %} {%- if not loop.first %} {{- raise_exception('System message must be at the beginning.') }} {%- endif %} {%- elif message.role == "user" %} {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }} {%- elif message.role == "assistant" %} {%- set reasoning_content = '' %} {%- if message.reasoning_content is string %} {%- set reasoning_content = message.reasoning_content %} {%- else %} {%- if '</think>' in content %} {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %} {%- set content = content.split('</think>')[-1].lstrip('\n') %} {%- endif %} {%- endif %} {%- set reasoning_content = reasoning_content|trim %} {%- if loop.index0 > ns.last_query_index %} {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content + '\n</think>\n\n' + content }} {%- else %} {{- '<|im_start|>' + message.role + '\n' + content }} {%- endif %} {%- if message.tool_calls and message.tool_calls is iterable and message.tool_calls is not mapping %} {%- for tool_call in message.tool_calls %} {%- if tool_call.function is defined %} {%- set tool_call = tool_call.function %} {%- endif %} {%- if loop.first %} {%- if content|trim %} {{- '\n\n<tool_call>\n<function=' + tool_call.name + '>\n' }} {%- else %} {{- '<tool_call>\n<function=' + tool_call.name + '>\n' }} {%- endif %} {%- else %} {{- '\n<tool_call>\n<function=' + tool_call.name + '>\n' }} {%- endif %} {%- if tool_call.arguments is defined %} {%- for args_name, args_value in tool_call.arguments|items %} {{- '<parameter=' + args_name + '>\n' }} {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string %} {{- args_value }} {{- '\n</parameter>\n' }} {%- endfor %} {%- endif %} {{- '</function>\n</tool_call>' }} {%- endfor %} {%- endif %} {{- '<|im_end|>\n' }} {%- elif message.role == "tool" %} {%- if loop.previtem and loop.previtem.role != "tool" %} {{- '<|im_start|>user' }} {%- endif %} {{- '\n<tool_response>\n' }} {{- content }} {{- '\n</tool_response>' }} {%- if not loop.last and loop.nextitem.role != "tool" %} {{- '<|im_end|>\n' }} {%- elif loop.last %} {{- '<|im_end|>\n' }} {%- endif %} {%- else %} {{- raise_exception('Unexpected message role.') }} {%- endif %} {%- endfor %} {%- if add_generation_prompt %} {{- '<|im_start|>assistant\n' }} {%- if enable_thinking is defined and enable_thinking is false %} {{- '<think>\n\n</think>\n\n' }} {%- else %} {{- '<think>\n' }} {%- endif %} {%- endif %}

Ratings by subcategory

Metric: Grade Norm
Model, team 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 8_0 8_1 8_2 8_3 8_4
Qwen3.5-4B
MERA
0.2 0.133 0.8 0.033 0.033 0.1 0.033 - 0 0 0 0.033 0.067 0.067 0 0.067 0.033 0 0.133 0.033 0.233 0.433 0.4 0.1 0.067 0.492 0.1 0.2 0.2 0.233 0.233
Model, team Honest Helpful Harmless
Qwen3.5-4B
MERA
0.607 0.695 0.707
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Management Philosophy Prehistory Human aging Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Biology (college) Physics (college) Human Sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine learning Medical genetics Professional law PR Security studies Chemistry (школьная) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual_physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) European history Government and politics
Qwen3.5-4B
MERA
0.711 0.464 0.849 0.799 0.699 0.776 0.777 0.691 0.719 0.596 0.667 0.77 0.45 0.704 0.791 0.584 0.71 0.819 0.944 0.756 0.554 0.772 0.86 0.705 0.652 0.9 0.354 0.639 0.669 0.68 0.76 0.769 0.718 0.747 0.721 0.863 0.85 0.874 0.861 0.857 0.747 0.743 0.772 0.926 0.826 0.759 0.711 0.933 0.635 0.69 0.78 0.709 0.856 0.887 0.83 0.752 0.85
Model, team SIM FL STA
Qwen3.5-4B
MERA
0.113 0.555 0.903
Model, team Anatomy Virology Astronomy Marketing Nutrition Sociology Managment Philosophy Pre-History Gerontology Econometrics Formal logic Global facts Jurisprudence Miscellaneous Moral disputes Business ethics Bilology (college) Physics (college) Human sexuality Moral scenarios World religions Abstract algebra Medicine (college) Machine Learning Genetics Professional law PR Security Chemistry (college) Computer security International law Logical fallacies Politics Clinical knowledge Conceptual physics Math (college) Biology (high school) Physics (high school) Chemistry (high school) Geography (high school) Professional medicine Electrical Engineering Elementary mathematics Psychology (high school) Statistics (high school) History (high school) Math (high school) Professional Accounting Professional psychology Computer science (college) World history (high school) Macroeconomics Microeconomics Computer science (high school) Europe History Government and politics
Qwen3.5-4B
MERA
0.622 0.822 0.75 0.713 0.855 0.724 0.707 0.754 0.769 0.738 0.808 0.825 0.542 0.76 0.708 0.691 0.692 0.778 0.912 0.702 0.667 0.831 0.867 0.917 0.911 0.833 0.821 0.667 0.912 0.889 0.778 0.872 0.75 0.825 0.652 0.821 0.889 0.844 0.877 0.892 0.804 0.873 0.733 0.978 0.81 0.889 0.897 0.864 0.785 0.86 0.844 0.797 0.759 0.74 0.744 0.637 0.911
Coorect
Good
Ethical
Model, team Virtue Law Moral Justice Utilitarianism
Qwen3.5-4B
MERA
0.22 0.229 0.231 0.194 0.154
Model, team Virtue Law Moral Justice Utilitarianism
Qwen3.5-4B
MERA
0.298 0.289 0.316 0.285 0.249
Model, team Virtue Law Moral Justice Utilitarianism
Qwen3.5-4B
MERA
0.283 0.278 0.29 0.265 0.232
Model, team Women Men LGBT Nationalities Migrants Other
Qwen3.5-4B
MERA
0.593 0.429 0.471 0.568 0.429 0.59