DEEPSEEK V4 FLASH
284B parameters with 13B active (MoE). 1M token context. Hybrid reasoning toggle. $0.14 input / $0.28 output per million tokens. The replacement for deepseek-chat.
DeepSeek V4 Flash is the cheap-fast variant of DeepSeek's V4 family. It's a 284B-parameter mixture-of-experts model that activates only 13B parameters per token, giving it frontier-adjacent quality at a fraction of the inference cost. At $0.14 per million input tokens and $0.28 per million output, it is roughly 35x cheaper on input and 100x cheaper on output than GPT-5.5 ($5/$30) — and it ships a 1M-token context window in production. The hybrid reasoning toggle lets callers flip between fast non-thinking responses and a deeper chain-of-thought mode when the prompt demands it. The case for: bulk classification, summarization, codegen at scale, RAG pipelines, anything where the marginal token cost matters. It officially replaces the older deepseek-chat endpoint. The case against: it lags Claude Opus 4.7 and GPT-5.5 on the hardest reasoning and refactor tasks, and tool-calling reliability still trails OpenAI in long agent loops.
Against the frontier lineup, V4 Flash is the volume play — not the quality ceiling.
No — it replaces deepseek-chat. V4 Flash is the new cheap-fast endpoint in the V4 family, with a 1M-token context and the hybrid reasoning toggle.
$0.14 per million input tokens and $0.28 per million output tokens. It is the cheapest large-context model in the Council AI lineup.
It's a mixture-of-experts (MoE) design. The model has 284B total parameters but only routes 13B of them per token. You get the knowledge breadth of a large model at the inference cost of a small one.
Pick V4 Flash for bulk work, RAG pipelines, and price-sensitive codegen. Pick V4 Pro when you need the harder reasoning ceiling.
Yes, but tool-calling stability in long agent loops still trails GPT-5.5. For short tool sequences and structured outputs it's reliable.
Yes — Council AI is fully managed and ships V4 Flash on every tier.