Qwen3-0.6B lineage
Hub tags list Qwen/Qwen3-0.6B-Base as base model and fine-tune source, and the model card gives a Qwen3-0.6B overview with 0.6B parameters and pretraining plus post-training stages.
Open Source Model Profile · Qwen
Qwen3-0.6B-MLX-bf16
Qwen3-0.6B-MLX-bf16 is a 0.6B-parameter Qwen3 text-generation release from Qwen for MLX. According to the model card, it supports switching between thinking and non-thinking modes.
- Publisher
- Qwen
- Task
- text-generation
- Model type
- qwen3
- License
- apache-2.0
- Library
- mlx
- Publication status
- Accepted · not indexed
Model overview
Qwen3-0.6B-MLX-bf16 is published by Qwen as a Qwen3 text-generation model. The captured configuration identifies Qwen3ForCausalLM and Safetensors metadata reports 596049920 parameters. Hub tags list Qwen3-0.6B-Base as base model, the card overview lists 0.6B parameters with pretraining and post-training stages, and card data records apache-2.0.
Recorded capabilities
Thinking mode switching
According to the model card, the model supports thinking and non-thinking modes with an enable_thinking switch and /think and /no_think turn-by-turn controls.
32k context and GQA detail
According to the model card, Qwen3-0.6B lists 32,768 context length, 28 layers, and 16-for-Q with 8-for-KV attention heads.
MLX BF16 release
The record is tagged with the mlx library, and the model card documents loading this BF16 release through the MLX load workflow with chat-template support.
Documented sampling guidance
According to the model card, thinking-mode guidance lists Temperature 0.6, TopP 0.95, TopK 20, and MinP 0, with boxed math formatting and JSON choice formatting for specific tasks.
Use cases in the source record
- Conversational text generation using the documented MLX load and chat-template workflow.
- Reasoning-mode experiments that use the publisher-documented thinking switch and /think and /no_think controls.
- Tool-integration experiments described in the card's agent and qwen-agent Assistant examples, treated as publisher guidance rather than verified capability.
Limitations and unknowns
- No independent evaluation results were extracted; publisher comparisons to QwQ and Qwen2.5 instruct models are untested publisher claims.
- No pricing, VRAM, throughput, or hardware-requirement values were extracted from this record.
- Provider state is historical snapshot data, not independently refreshed current availability.
- Capability descriptions for reasoning, alignment, and agent use come from the publisher model card and were not independently verified.
Source and provenance
Source: Qwen/Qwen3-0.6B-MLX-bf16
Captured: Unknown. Processed: 2026-09-07T19:35:51.222324+00:00.
Qwen3-0.6B-MLX-bf16 Qwen3 Highlights Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features: Uniquely support of seamless switching between thinking mode (for complex logical reasoning, math, and coding) and non-thinking mode (for efficient, general-purpose dialogue) within single model , ensuring optimal performance across various scenarios. Significantl…
F001F002F003F004F005F006F007F008F009F010F011F012F013F014F015F016F017F018F020F021F022F023