Naive-N0.5-Flash
A 309-billion-parameter MoE open-source model with approximately 15.5 billion active parameters. It natively supports a one-million-token context window, targets code development and AI research domains, and is trained on top of Xiaomi's MiMo-V2.5.
Timeline (last 90 days)
Naive-N0.5-Flash is a 309B-parameter MoE (Mixture of Experts) model with approximately 15.5B activated parameters
Naive-N0.5-Flash natively supports a 1 million token context and does not use a full attention mechanism
Naive-N0.5-Flash can be optimized to approximately 2000 tokens/s in single-concurrency scenarios, and can exceed 50 tokens per second in multi-concurrency scenarios
Naive-N0.5-Flash was publicly trained based on Xiaomi's MiMo-V2.5 model
The official announcement claims that Naive-N0.5-Flash ranks just below some leading models on benchmarks such as SWE-bench 1.1, and claims leading performance among open-source models
The BF16 weight files for Naive-N0.5-Flash are approximately 618GB, and the FP8 format version weights are approximately 315GB
Naive-N0.5-Flash does not support multimodality
Before deletion, Naive-N0.5-Flash provided a reproducible physical environment and complete toolchain, but the toolchain was subsequently deleted
MoE models only activate a small subset of expert sub-networks during each inference, and Naive-N0.5-Flash has only approximately 5% of its parameters working during actual inference
UP host Kunmengtou believes that Naive-N0.5-Flash is more of a conceptual marketing exercise than a genuine technological breakthrough
All Facts (9)
Naive-N0.5-Flash is a 309B-parameter MoE (Mixture of Experts) model with approximately 15.5B activated parameters
50%UnverifiedNaive-N0.5-Flash natively supports a 1 million token context and does not use a full attention mechanism
50%UnverifiedNaive-N0.5-Flash can be optimized to approximately 2000 tokens/s in single-concurrency scenarios, and can exceed 50 tokens per second in multi-concurrency scenarios
50%UnverifiedNaive-N0.5-Flash was publicly trained based on Xiaomi's MiMo-V2.5 model
50%UnverifiedThe official announcement claims that Naive-N0.5-Flash ranks just below some leading models on benchmarks such as SWE-bench 1.1, and claims leading performance among open-source models
50%UnverifiedThe BF16 weight files for Naive-N0.5-Flash are approximately 618GB, and the FP8 format version weights are approximately 315GB
50%UnverifiedNaive-N0.5-Flash does not support multimodality
50%UnverifiedMoE models only activate a small subset of expert sub-networks during each inference, and Naive-N0.5-Flash has only approximately 5% of its parameters working during actual inference
50%UnverifiedUP host Kunmengtou believes that Naive-N0.5-Flash is more of a conceptual marketing exercise than a genuine technological breakthrough
50%