GoKawiilDeepSeek updates V4.1 Flash model to revision 1017, cuts memory use sharply
DeepSeek released a point update to its open-weight V4.1 Flash model, a 552B-parameter Mixture-of-Experts system with 16B active parameters and a 1-million-token context window that handles both text and images. The company says the new revision needs only a quarter of the HBM and an eighth of the SSD storage for its KV cache compared with the prior version, using 890 bytes per token entry, and its internal benchmarks place the model between GPT-5.6 Sol and Claude Opus 5, beating the latter on some task tests.
GoKawiil's interpretation of the reporting above, not reported fact.
The efficiency gains suggest DeepSeek is pushing to close the cost gap with Western frontier labs while keeping the openness and low running costs that have made Chinese open-weight models attractive for agentic workloads. Lower memory and storage demands could make long-running, multi-step AI tasks cheaper to operate at scale, though the model's reported tendency toward hallucination may limit its use in reliability-critical settings.
- DeepSeek V4.1 Flash revision 1017 cuts KV cache memory use to a quarter of HBM and an eighth of SSD storage versus its predecessor.
- The model is a 552B-parameter MoE system with 16B active parameters, multimodal input, and a 1-million-token context window.
- DeepSeek's own benchmarks place it between GPT-5.6 Sol and Claude Opus 5 in performance, despite concerns about hallucination-prone output.
Source: tomshardware.com — Bruno Ferreira, 2026-10-23
Published there as: “Deepseek's latest AI model is smarter than ever, but its low running costs come at a price”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.