TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill

Qwen3.5 27B x Claude Opus 4.6

Big thanks to @EclipseMist for providing the LoRAs for this model

• 🧬 Datasets:

  • crownelius/Opus-4.6-Reasoning-2100x-formatted
  • Personal Claude Data provided by @EclipseMist

• 🏗 Base Model:

  • unsloth/Qwen3.5-27B

• ⚡ Use cases:

  • Coding
  • Creative Writing
  • Visual Understanding
  • General Purpose

Citations and Contributions

• @EclipseMist - Training and Data Curation
• @crownelius - Data Curation
• @unsloth - This qwen3 model was trained 2x faster with Unsloth and Huggingface's TRL library.
• @Qwen - Providing a fantastic, native-multimodal base model

---

Benchmarks

Benchmark	TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	unsloth/Qwen3.5-27B
arc_challenge	0.461	0.435
gpqa_diamond_zeroshot	0.283	0.283
hellaswag	0.613	0.574
mmlu	0.233	0.230
truthfulqa_mc2	0.610	0.599
winogrande	0.769	0.749

Table

Model	Benchmark	Score	Total Questions	Total Correct
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	arc_challenge	0.460751	1172	540
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	gpqa_diamond_zeroshot	0.282828	198	56
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	hellaswag	0.612926	10042	6155
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	mmlu	0.232944	14042	3271
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	truthfulqa_mc2	0.610146	817	498
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	winogrande	0.768745	1267	974
unsloth/Qwen3.5-27B	arc_challenge	0.435154	1172	510
unsloth/Qwen3.5-27B	gpqa_diamond_zeroshot	0.282828	198	56
unsloth/Qwen3.5-27B	hellaswag	0.574288	10042	5767
unsloth/Qwen3.5-27B	mmlu	0.229597	14042	3224
unsloth/Qwen3.5-27B	truthfulqa_mc2	0.599243	817	489
unsloth/Qwen3.5-27B	winogrande	0.749013	1267	949

MMLU Subject Breakdown

Table

Model	Subject	Benchmark	Score	Total Questions	Total Correct
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	formal_logic	mmlu_formal_logic	0.285714	126	36
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_european_history	mmlu_high_school_european_history	0.224242	165	37
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_us_history	mmlu_high_school_us_history	0.240196	204	49
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_world_history	mmlu_high_school_world_history	0.274262	237	65
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	international_law	mmlu_international_law	0.239669	121	29
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	jurisprudence	mmlu_jurisprudence	0.268519	108	29
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	logical_fallacies	mmlu_logical_fallacies	0.226994	163	37
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	moral_disputes	mmlu_moral_disputes	0.263006	346	91
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	moral_scenarios	mmlu_moral_scenarios	0.237989	895	213
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	philosophy	mmlu_philosophy	0.192926	311	59
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	prehistory	mmlu_prehistory	0.209877	324	68
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	professional_law	mmlu_professional_law	0.245111	1534	376
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	world_religions	mmlu_world_religions	0.321637	171	55
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	business_ethics	mmlu_business_ethics	0.3	100	30
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	clinical_knowledge	mmlu_clinical_knowledge	0.222642	265	59
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_medicine	mmlu_college_medicine	0.208092	173	36
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	global_facts	mmlu_global_facts	0.19	100	19
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	human_aging	mmlu_human_aging	0.313901	223	70
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	management	mmlu_management	0.194175	103	20
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	marketing	mmlu_marketing	0.290598	234	68
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	medical_genetics	mmlu_medical_genetics	0.29	100	28
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	miscellaneous	mmlu_miscellaneous	0.259259	783	203
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	nutrition	mmlu_nutrition	0.222222	306	68
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	professional_accounting	mmlu_professional_accounting	0.230496	282	65
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	professional_medicine	mmlu_professional_medicine	0.183824	272	50
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	virology	mmlu_virology	0.283133	166	47
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	econometrics	mmlu_econometrics	0.236842	114	27
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_geography	mmlu_high_school_geography	0.171717	198	34
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_government_and_politics	mmlu_high_school_government_and_politics	0.196891	193	38
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_macroeconomics	mmlu_high_school_macroeconomics	0.205128	390	80
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_microeconomics	mmlu_high_school_microeconomics	0.210084	238	50
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_psychology	mmlu_high_school_psychology	0.201835	545	110
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	human_sexuality	mmlu_human_sexuality	0.259542	131	34
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	professional_psychology	mmlu_professional_psychology	0.25817	612	158
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	public_relations	mmlu_public_relations	0.236364	110	26
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	security_studies	mmlu_security_studies	0.191837	245	47
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	sociology	mmlu_sociology	0.268657	201	54
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	us_foreign_policy	mmlu_us_foreign_policy	0.27	100	27
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	abstract_algebra	mmlu_abstract_algebra	0.22	100	22
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	anatomy	mmlu_anatomy	0.2	135	27
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	astronomy	mmlu_astronomy	0.177632	152	27
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_biology	mmlu_college_biology	0.263889	144	38
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_chemistry	mmlu_college_chemistry	0.19	100	19
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_computer_science	mmlu_college_computer_science	0.26	100	26
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_mathematics	mmlu_college_mathematics	0.21	100	21
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	college_physics	mmlu_college_physics	0.215686	102	22
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	computer_security	mmlu_computer_security	0.28	100	28
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	conceptual_physics	mmlu_conceptual_physics	0.26383	235	62
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	electrical_engineering	mmlu_electrical_engineering	0.241379	145	35
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	elementary_mathematics	mmlu_elementary_mathematics	0.21164	378	80
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_biology	mmlu_high_school_biology	0.190323	310	58
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_chemistry	mmlu_high_school_chemistry	0.152709	203	31
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_computer_science	mmlu_high_school_computer_science	0.25	100	25
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_mathematics	mmlu_high_school_mathematics	0.211111	270	57
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_physics	mmlu_high_school_physics	0.198675	151	29
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	high_school_statistics	mmlu_high_school_statistics	0.152778	216	33
TeichAI/Qwen3.5-27B-Claude-Opus-4.6-Distill	machine_learning	mmlu_machine_learning	0.3125	112	35
unsloth/Qwen3.5-27B	formal_logic	mmlu_formal_logic	0.285714	126	36
unsloth/Qwen3.5-27B	high_school_european_history	mmlu_high_school_european_history	0.218182	165	36
unsloth/Qwen3.5-27B	high_school_us_history	mmlu_high_school_us_history	0.25	204	51
unsloth/Qwen3.5-27B	high_school_world_history	mmlu_high_school_world_history	0.270042	237	63
unsloth/Qwen3.5-27B	international_law	mmlu_international_law	0.239669	121	29
unsloth/Qwen3.5-27B	jurisprudence	mmlu_jurisprudence	0.259259	108	28
unsloth/Qwen3.5-27B	logical_fallacies	mmlu_logical_fallacies	0.220859	163	36
unsloth/Qwen3.5-27B	moral_disputes	mmlu_moral_disputes	0.248555	346	86
unsloth/Qwen3.5-27B	moral_scenarios	mmlu_moral_scenarios	0.237989	895	213
unsloth/Qwen3.5-27B	philosophy	mmlu_philosophy	0.186495	311	58
unsloth/Qwen3.5-27B	prehistory	mmlu_prehistory	0.216049	324	70
unsloth/Qwen3.5-27B	professional_law	mmlu_professional_law	0.245763	1534	377
unsloth/Qwen3.5-27B	world_religions	mmlu_world_religions	0.321637	171	55
unsloth/Qwen3.5-27B	business_ethics	mmlu_business_ethics	0.3	100	30
unsloth/Qwen3.5-27B	clinical_knowledge	mmlu_clinical_knowledge	0.215094	265	57
unsloth/Qwen3.5-27B	college_medicine	mmlu_college_medicine	0.208092	173	36
unsloth/Qwen3.5-27B	global_facts	mmlu_global_facts	0.18	100	18
unsloth/Qwen3.5-27B	human_aging	mmlu_human_aging	0.313901	223	70
unsloth/Qwen3.5-27B	management	mmlu_management	0.174757	103	18
unsloth/Qwen3.5-27B	marketing	mmlu_marketing	0.290598	234	68
unsloth/Qwen3.5-27B	medical_genetics	mmlu_medical_genetics	0.3	100	30
unsloth/Qwen3.5-27B	miscellaneous	mmlu_miscellaneous	0.240102	783	188
unsloth/Qwen3.5-27B	nutrition	mmlu_nutrition	0.22549	306	69
unsloth/Qwen3.5-27B	professional_accounting	mmlu_professional_accounting	0.234043	282	66
unsloth/Qwen3.5-27B	professional_medicine	mmlu_professional_medicine	0.183824	272	50
unsloth/Qwen3.5-27B	virology	mmlu_virology	0.283133	166	47
unsloth/Qwen3.5-27B	econometrics	mmlu_econometrics	0.236842	114	27
unsloth/Qwen3.5-27B	high_school_geography	mmlu_high_school_geography	0.176768	198	35
unsloth/Qwen3.5-27B	high_school_government_and_politics	mmlu_high_school_government_and_politics	0.196891	193	38
unsloth/Qwen3.5-27B	high_school_macroeconomics	mmlu_high_school_macroeconomics	0.202564	390	79
unsloth/Qwen3.5-27B	high_school_microeconomics	mmlu_high_school_microeconomics	0.210084	238	50
unsloth/Qwen3.5-27B	high_school_psychology	mmlu_high_school_psychology	0.192661	545	105
unsloth/Qwen3.5-27B	human_sexuality	mmlu_human_sexuality	0.259542	131	34
unsloth/Qwen3.5-27B	professional_psychology	mmlu_professional_psychology	0.25	612	153
unsloth/Qwen3.5-27B	public_relations	mmlu_public_relations	0.218182	110	24
unsloth/Qwen3.5-27B	security_studies	mmlu_security_studies	0.187755	245	46
unsloth/Qwen3.5-27B	sociology	mmlu_sociology	0.243781	201	49
unsloth/Qwen3.5-27B	us_foreign_policy	mmlu_us_foreign_policy	0.28	100	28
unsloth/Qwen3.5-27B	abstract_algebra	mmlu_abstract_algebra	0.22	100	22
unsloth/Qwen3.5-27B	anatomy	mmlu_anatomy	0.185185	135	25
unsloth/Qwen3.5-27B	astronomy	mmlu_astronomy	0.177632	152	27
unsloth/Qwen3.5-27B	college_biology	mmlu_college_biology	0.256944	144	37
unsloth/Qwen3.5-27B	college_chemistry	mmlu_college_chemistry	0.2	100	20
unsloth/Qwen3.5-27B	college_computer_science	mmlu_college_computer_science	0.26	100	26
unsloth/Qwen3.5-27B	college_mathematics	mmlu_college_mathematics	0.21	100	21
unsloth/Qwen3.5-27B	college_physics	mmlu_college_physics	0.215686	102	22
unsloth/Qwen3.5-27B	computer_security	mmlu_computer_security	0.28	100	28
unsloth/Qwen3.5-27B	conceptual_physics	mmlu_conceptual_physics	0.26383	235	62
unsloth/Qwen3.5-27B	electrical_engineering	mmlu_electrical_engineering	0.241379	145	35
unsloth/Qwen3.5-27B	elementary_mathematics	mmlu_elementary_mathematics	0.208995	378	79
unsloth/Qwen3.5-27B	high_school_biology	mmlu_high_school_biology	0.177419	310	55
unsloth/Qwen3.5-27B	high_school_chemistry	mmlu_high_school_chemistry	0.152709	203	31
unsloth/Qwen3.5-27B	high_school_computer_science	mmlu_high_school_computer_science	0.25	100	25
unsloth/Qwen3.5-27B	high_school_mathematics	mmlu_high_school_mathematics	0.211111	270	57
unsloth/Qwen3.5-27B	high_school_physics	mmlu_high_school_physics	0.198675	151	29
unsloth/Qwen3.5-27B	high_school_statistics	mmlu_high_school_statistics	0.152778	216	33
unsloth/Qwen3.5-27B	machine_learning	mmlu_machine_learning	0.3125	112	35

---

The following best practices recommended by Qwen

Best Practices

To achieve optimal performance, we recommend the following settings:

1. Sampling Parameters:

   • We suggest using the following sets of sampling parameters depending on the mode and task type:
     • Thinking mode for general tasks:
       temperature=1.0, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
     • Thinking mode for precise coding tasks (e.g., WebDev):
       temperature=0.6, top_p=0.95, top_k=20, min_p=0.0, presence_penalty=0.0, repetition_penalty=1.0
     • Instruct (or non-thinking) mode for general tasks:
       temperature=0.7, top_p=0.8, top_k=20, min_p=0.0, presence_penalty=1.5, repetition_penalty=1.0
     • Instruct (or non-thinking) mode for reasoning tasks:
       temperature=1.0, top_p=1.0, top_k=40, min_p=0.0, presence_penalty=2.0, repetition_penalty=1.0
   • For supported frameworks, you can adjust the presence_penalty parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.

2. Adequate Output Length: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 81,920 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.

3. Standardize Output Format: We recommend using prompts to standardize model outputs when benchmarking.

   • Math Problems: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
   • Multiple-Choice Questions: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the answer field with only the choice letter, e.g., "answer": "C"."

4. No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.

5. Long Video Understanding: To optimize inference efficiency for plain text and images, the size parameter in the released video_preprocessor_config.json is conservatively configured. It is recommended to set the longest_edge parameter in the video_preprocessor_config file to 469,762,048 (corresponding to 224k video tokens) to enable higher frame-rate sampling for hour-scale videos and thereby achieve superior performance. For example,

   {"longest_edge": 469762048, "shortest_edge": 4096}
   

   Alternatively, override the default values via engine startup parameters. For implementation details, refer to: vLLM / SGLang.