DeepSeekAI ModelsLLM PricingAI BenchmarksAI Agents

DeepSeek V4 Flash: Current API Name, Pricing, and Model Size

DeepSeek's API now serves V4.1-Flash through the deepseek-flash name. Review the current peak and off-peak price schedule, context size, and the limits of active-parameter comparisons.

Published: Last checked:

DeepSeek V4 Flash model and pricing analysis
Analyzed Foundation Models (4)All Models Directory→

DeepSeek’s current API documentation identifies the Flash model as DeepSeek-V4.1-Flash and says to call it using deepseek-flash. The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp remain accepted for compatibility but route to V4.1-Flash. This is distinct from older V4 Flash 0731 model specifications and prices.

DeepSeek’s pricing page lists separate peak and off-peak rates. Its September announcement describes the new V4.1-Flash model and pricing schedule. These are API facts; third-party hosts may use different model snapshots, rate cards, and serving paths.

Current API specifications

ItemCurrent documented detail
API model namedeepseek-flash
Model versionDeepSeek-V4.1-Flash
Context lengthUp to 1M tokens
Maximum outputUp to 384K tokens
Architecture552B-parameter MoE
Active parameters8B for input; 16B for output, per DeepSeek’s report
FeaturesThinking and non-thinking modes; tool calls and JSON output are listed in API docs

The active parameter count should not be confused with model size or hardware memory. For example, 8B active parameters do not imply that an 8B-weight-sized GPU can store the model. The technical report describes a 552B-parameter model; serving also requires memory for runtime state, context, and caches. Local deployment depends on supported inference software and quantization and hardware configurations, so do not infer that a consumer dual-GPU setup can run the full model from the active count alone.

Current native API prices

Prices are per one million tokens:

Token typeOff-peakPeak
Input, cache miss$0.15$0.30
Input, cache hit$0.003$0.006
Output$0.60$1.20

DeepSeek states that off-peak rates are half of peak rates and publishes the UTC schedule on its pricing page. It also reserves the right to change prices. A simple estimate should separate uncached input, cached input, and output instead of multiplying all tokens by one blended rate. For illustration, a million cache-miss input tokens and a million output tokens would cost $0.75 off-peak or $1.50 at peak, before any other billing consideration.

What the architecture figures tell you

The V4.1-Flash report describes an asymmetric design with different active parameter counts for input and output, a one-million-token context, and compressed KV-cache needs relative to the previous generation. The report’s comparisons are tied to its defined baseline and technical setup. They do not show that every workload sees the same speed or memory reduction, and they are not direct benchmark guarantees.

For API users, the practical advantages to evaluate are the documented context size, supported modes, current prices, and performance on representative prompts. Measure latency and answer quality with your actual request shape, including cache behavior and generated output length. A maximum context window is not a recommendation to send every task at that size.

Takeaway

For DeepSeek’s current Flash API, use deepseek-flash and check the peak/off-peak rate card before calculating cost. Legacy Flash IDs route to V4.1-Flash, whose reported size is much larger than the active parameter count might suggest. Hosted-provider pricing and older model cards should be treated separately from DeepSeek’s native API terms.

Frequently Asked Questions

What model does DeepSeek's Flash API name use now?

DeepSeek's pricing docs say to use `deepseek-flash` for V4.1-Flash. The legacy `deepseek-v4-flash` and `deepseek-v4-flash-vision-exp` names are still accepted for compatibility but route to V4.1-Flash.

How much does DeepSeek V4.1-Flash cost?

The current official API rate card lists off-peak input at $0.15 per million cache misses and $0.003 per million cache hits, plus $0.60 per million output tokens. Peak prices are $0.30, $0.006, and $1.20 respectively.

How large is DeepSeek V4.1-Flash?

DeepSeek's September 2026 report describes a 552-billion-parameter MoE with 8 billion active parameters for input and 16 billion for output, and context up to one million tokens.

Does 8B active parameters mean V4.1-Flash fits in 8B worth of GPU memory?

No. Active parameter count describes the portion used in a token computation; storing model weights requires memory for the full model, plus runtime and context memory.

Sources & disclosures

Source links for data and claims discussed in this article are listed below. Refer to each source for its current figures and methodology.

Share Article

Recommended Reading

Related by topic and recency.

All articles→
NewsDeepSeek

DeepSeek V4 Flash 0731: Current API Alias and Historical Context

DeepSeek's API now serves legacy V4 Flash aliases as V4.1-Flash. See the current context limits and peak/off-peak prices, and why old snapshot benchmarks need separate evidence.

Aug 10, 2026 · 2 min read
ReviewsAI Models

Alibaba Model Studio Token Plan: Pricing Structure and Claude Code Integration

Review of Alibaba Model Studio Token Plan terms, Singapore availability, Personal and Team quotas, API compatibility, and billing checks before subscribing.

Sep 18, 2026 · 3 min read
NewsAI Models

Tencent Hy4 Preview: 770B MoE Architecture and 1M Context Window

Tencent Hy4 preview has 770B total and 49B activated parameters, a 1M context window, published FP8 checkpoints, deployment recipes, and an Apache 2.0 license.

Aug 29, 2026 · 3 min read