[ {"id":"qwen3_0p6b","family":"qwen","repo":"Qwen/Qwen3-0.6B","requested":"Qwen/Qwen3-0.6B","params_b":0.6,"reasoning":false}, {"id":"qwen3_1p7b","family":"qwen","repo":"Qwen/Qwen3-1.7B","requested":"Qwen/Qwen3-1.7B","params_b":1.7,"reasoning":false}, {"id":"qwen3_4b","family":"qwen","repo":"Qwen/Qwen3-4B","requested":"Qwen/Qwen3-4B","params_b":4.0,"reasoning":false}, {"id":"qwen3_8b","family":"qwen","repo":"Qwen/Qwen3-8B","requested":"Qwen/Qwen3-8B","params_b":8.2,"reasoning":false}, {"id":"qwen3_14b","family":"qwen","repo":"Qwen/Qwen3-14B","requested":"Qwen/Qwen3-14B","params_b":14.8,"reasoning":false}, {"id":"qwen3_32b","family":"qwen","repo":"Qwen/Qwen3-32B","requested":"Qwen/Qwen3-32B","params_b":32.8,"reasoning":false}, {"id":"r1_distill_qwen_7b","family":"deepseek","repo":"deepseek-ai/DeepSeek-R1-Distill-Qwen-7B","requested":"deepseek-ai/DeepSeek-R1-Distill-Qwen-7B","params_b":7.6,"reasoning":true}, {"id":"r1_distill_qwen_14b","family":"deepseek","repo":"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B","requested":"deepseek-ai/DeepSeek-R1-Distill-Qwen-14B","params_b":14.8,"reasoning":true}, {"id":"ministral3_8b","family":"mistral","repo":"mistralai/Ministral-3-8B-Instruct-2512","requested":"current efficient ~8B Mistral instruct","params_b":8.0,"reasoning":false}, {"id":"magistral_small_24b","family":"mistral","repo":"mistralai/Magistral-Small-2509","requested":"Mistral reasoning/stronger 14-24B","params_b":24.0,"reasoning":true,"fallback":"mistralai/Ministral-3-14B-Instruct-2512"}, {"id":"gemma3_4b","family":"gemma","repo":"unsloth/gemma-3-4b-it","requested":"google/gemma-3-4b-it","params_b":4.3,"reasoning":false,"reason":"official repo is manually gated and no HF token is available; unsloth mirror hosts the same weights (Gemma license applies)"}, {"id":"gemma3_12b","family":"gemma","repo":"unsloth/gemma-3-12b-it","requested":"google/gemma-3-12b-it","params_b":12.2,"reasoning":false,"reason":"same as above"}, {"id":"phi4_mini_reasoning","family":"phi","repo":"microsoft/Phi-4-mini-reasoning","requested":"microsoft/Phi-4-mini-reasoning","params_b":3.8,"reasoning":true}, {"id":"phi4","family":"phi","repo":"microsoft/phi-4","requested":"microsoft/phi-4","params_b":14.7,"reasoning":false}, {"id":"granite33_2b","family":"granite","repo":"ibm-granite/granite-3.3-2b-instruct","requested":"ibm-granite/granite-3.3-2b-instruct","params_b":2.5,"reasoning":false}, {"id":"granite33_8b","family":"granite","repo":"ibm-granite/granite-3.3-8b-instruct","requested":"ibm-granite/granite-3.3-8b-instruct","params_b":8.2,"reasoning":false}, {"id":"llama31_8b","family":"llama","repo":"unsloth/Llama-3.1-8B-Instruct","requested":"meta-llama/Llama-3.1-8B-Instruct","params_b":8.0,"reasoning":false,"reason":"official repo manually gated, no HF token; unsloth mirror hosts the same weights (Llama 3.1 community license applies)"}, {"id":"llama32_3b","family":"llama","repo":"unsloth/Llama-3.2-3B-Instruct","requested":"another accessible Llama instruct in a different size tier","params_b":3.2,"reasoning":false,"reason":"Llama-3.3-70B would need 2 GPUs and ~5x the compute; 3B gives a distinct small tier"}, {"id":"gpt_oss_20b","family":"openai","repo":"openai/gpt-oss-20b","requested":"openai/gpt-oss-20b","params_b":20.9,"active_params_b":3.6,"reasoning":true}, {"id":"olmo3_7b","family":"allenai","repo":"allenai/Olmo-3-7B-Instruct","requested":"additional strong open model from an underrepresented family","params_b":7.3,"reasoning":false,"fallback":"allenai/OLMo-2-1124-13B-Instruct"} ]