DeepSeek Releases V4-Flash-Base: 292B Parameter Base Model
DeepSeek has released V4-Flash-Base, a 292 billion parameter base model now available on Hugging Face. The model uses BF16, I64, F32, and F8_E4M3 tensor types and is distributed in Safetensors format.
DeepSeek V4-Flash-Base: 292B Parameter Base Model Released
DeepSeek has released V4-Flash-Base, a 292 billion parameter base model now available on Hugging Face. The model represents the base version of DeepSeek's V4-Flash series.
Technical Specifications
The model contains 292 billion parameters and supports multiple tensor types: BF16 (bfloat16), I64 (64-bit integer), F32 (32-bit float), and F8_E4M3 (8-bit float in E4M3 format). Files are distributed in the Safetensors format, which provides safer serialization than traditional pickle-based formats.
Availability and Deployment
The model weights are available for download on Hugging Face as part of a collection containing 4 items. According to the Hugging Face listing, no inference providers currently support deployment of this model. The collection was last updated approximately 4 hours ago and has 307 downloads.
Missing Information
DeepSeek has not yet published a model card with detailed information about training data, benchmark performance, capabilities, or pricing. Context window size, training cutoff date, and specific use cases remain undisclosed. As a base model, V4-Flash-Base typically requires fine-tuning for specific tasks, unlike instruction-tuned variants.
What This Means
The release of a 292B parameter base model signals DeepSeek's continued development of large-scale models, though the lack of documentation makes technical evaluation impossible at this stage. The "Flash" designation suggests optimization for speed, consistent with other models in the industry using similar naming conventions. The use of F8_E4M3 tensor types indicates potential support for efficient inference through quantization. Without benchmark scores or a detailed model card, organizations should wait for complete documentation before considering deployment.
Related Articles
Deepseek open-sources DeepInfer, the Apache 2.0 inference engine behind its V4.1-Flash model
Deepseek has open-sourced DeepInfer, the internal inference framework that serves its V4.1-Flash model, under the Apache License 2.0. The company says it will also open-source code for future models and has released deepseek-recipe to reproduce its API's token processing in self-hosted setups.
MiniMax M3.1 Flash scores 42 on AA Intelligence Index at $0.15/$0.60 per 1M tokens, open weights Oct 23
MiniMax has released M3.1 Flash, a reasoning model that scores 42 on the Artificial Analysis Intelligence Index, up from 29 for M3. It costs $0.15 per 1M input tokens and $0.60 per 1M output tokens, and open weights are planned for October 23.
DeepSeek V4.1 Flash 1017 launches with 552B CED backbone, 1M context, $0.15/$0.60 per 1M tokens
DeepSeek has released V4.1 Flash 1017, an October 17, 2026 snapshot built on its new Causal Encoder-Decoder architecture. It uses a 552B-parameter sparse MoE backbone with a 1M-token context window, native image understanding, and pricing of $0.15 input / $0.60 output per 1M tokens on OpenRouter.
Recursive releases Flint, a 40B-parameter open coding model that runs on laptops under Apache 2.0
Recursive released Flint, a free, open-weight mixture-of-experts coding model with 40 billion total and 2.5 billion active parameters, under Apache 2.0. The company claims its Eureka Machine automated research system contributed to nearly every stage of development, and that Flint runs at about 131 tokens per second on an M4 Pro MacBook Pro.
Comments
Loading...