Skip to main content

DeepSeek-V4.1-Flash

Overview​

DeepSeek-V4.1-Flash is a native multimodal Mixture-of-Experts model released by DeepSeek on September 10, 2026. It combines a 552-billion-parameter backbone, a 1M-token context window, and a Causal Encoder-Decoder design that activates 8 billion parameters per token during prefill and 16 billion during decoding. On B.AI, use the model name DeepSeek-V4.1-Flash.

B.AI model routing

B.AI will progressively route requests made using the DeepSeek-V4-Flash and DeepSeek-V4-Flash-Vision-Exp model names to DeepSeek-V4.1-Flash. After routing takes effect, these requests are billed at the applicable DeepSeek-V4.1-Flash price.

DeepSeek-V4.1-Flash

The offer takes effect at 10:00 on October 4, 2026 (SGT, UTC+8).

From that time, the promotional price changes in step with DeepSeek's time-based pricing and is billed at 60% of the applicable Off-Peak or Peak standard price.

The pricing table below continues to show standard reference prices. Actual settlement and final billing are subject to the platform display.

Limited-Time Discount60% from Oct 4, 10:00 SGT

Key Features​

  • Causal Encoder-Decoder Architecture: Uses a 40-layer Transformer split into a 20-layer causal encoder and a 20-layer decoder. Projecting the decoder's global KV cache from the encoder output reduces active computation for input-heavy agent workloads.
  • Compressed Long-Context Inference: CSA2, hierarchical sparse indexing, FP4 KV caching, and SWA Bounded Replay reduce the persistent global KV cache for long-context inference.
  • Native Multimodal Training: Processes images and text jointly and produces text output. Its vision encoder and projector are trained with the language model rather than attached only during post-training.
  • Controllable Reasoning: Supports low, high, and max reasoning-effort levels through the hosted API, with thinking enabled at high by default.
  • Structured and Tool-Driven Output: Supports JSON output, tool calls, context caching, Chat Prefix Completion, and FIM in non-thinking mode.

Best Use Cases​

  • Long-Horizon Coding Agents: Repository-scale implementation, debugging, testing, and terminal work that benefit from long context and tool use.
  • Multimodal Software Workflows: Inspecting screenshots, rendered interfaces, diagrams, and tool-produced images while modifying code or other artifacts.
  • Document and Chart Analysis: Extracting and reasoning over screenshots, scanned documents, dashboards, and charts through native image-and-text input.
  • High-Volume Agent Systems: Input-heavy workflows where compressed KV cache, low active-parameter counts, context caching, and Flash-tier pricing help control operating costs.
  • Structured Automation: Classification, extraction, JSON-formatted responses, and multi-step tool orchestration.

Capabilities and Limitations​

CapabilityDescription
ReasoningSupports low, high, and max reasoning-effort levels. Thinking is enabled at high by default; the Responses API can disable it with none.
Coding and AgentsDesigned for long-horizon coding, repository work, terminal tasks, and multi-step agent workflows.
MultimodalAccepts text and image input and produces text output.
Context Window1M tokens.
Max Output384K tokens.
Tool UseSupports function calling, JSON output, context caching, image-bearing tool output through the Responses API, Chat Prefix Completion, and FIM Completion in non-thinking mode.

Known Limitations​

  • The hosted API maps reasoning controls to low, high, and max; compatible effort values may be mapped rather than preserved as distinct levels.
  • In thinking mode, temperature, presence_penalty, and frequency_penalty do not affect generation. Tool-calling conversations must return prior reasoning_content in subsequent requests.
  • Images are accepted only in supported user or tool-output positions. Images in system or assistant messages return an HTTP 400 error.
  • DeepSeek does not separately publish a maximum input-token limit, knowledge cutoff, or complete supported-language list.

Standard Pricing​

The token prices below are shown in USD per 1 million tokens; web search is billed in USD per use.

Billing PeriodInput
(USD / 1M Tokens)		Cache Write
(USD / 1M Tokens)		Cache Read
(USD / 1M Tokens)		Output
(USD / 1M Tokens)		Web Search
(USD / use)		Billing Notes
Off-Peak$0.15$0.15$0.003$0.60-Cache Write: 1x input; Cache Read: 0.02x input
Peak$0.30$0.30$0.006$1.20-Cache Write: 1x input; Cache Read: 0.02x input

Credits settlement: B.AI converts charges at 1 USD = 1,000,000 Credits and deducts Credits from the account balance.

Pricing note

The table shows the time-based standard reference price for DeepSeek-V4.1-Flash. Off-Peak prices are half of Peak prices. In Beijing Time (UTC+8), 09:00-12:00 and 14:00-18:00, Monday through Friday (excluding Chinese public holidays), are Peak periods; all other times, including weekends and Chinese public holidays, are Off-Peak periods. DeepSeek-V4.1-Flash usage in B.AI Chat is billed at Off-Peak rates. Prices shown in the documentation are B.AI standard reference prices for base billing purposes. B.AI may provide lower actual usage costs through promotions and account benefits.