DeepSeek V4.1 Flash API Pricing & Token Cost Guide (2026)
DeepSeek has officially released DeepSeek-V4.1-Flash, pairing native multimodal vision and a 1,048,576-token context window with disruptive API pricing starting at $0.15/1M input and $0.003/1M cached tokens. Here is the complete breakdown of pricing tiers, caching math, and how it compares to V4 Pro, V3, and GPT-4o mini.
- API Identifier: Use
deepseek-flashonhttps://api.deepseek.com. - Disruptive Baseline Pricing: $0.15 per 1M input tokens and $0.60 per 1M output tokens during off-peak hours.
- Unmatched Cache Hit Savings: Prompt caching hits cost just $0.003 per 1M tokens (a 98% reduction).
- Massive 1.05M Context: 1,048,576 tokens context window with 384k maximum output limit.
- V4 Pro Deprecation & Routing: V4 Pro is being retired; starting September 14, 2026, calls to
deepseek-v4-proautomatically resolve to V4.1 Flash at the lower price.
💡 Model your exact costs for DeepSeek V4.1 Flash side-by-side →
Open Cost CalculatorDeepSeek V4.1 Flash Official API Pricing
DeepSeek utilizes dynamic pricing split between off-peak and peak utilization windows. Off-peak encompasses nights, weekends, and high-capacity slots, while peak covers standard UTC business hours. Even at peak rates, V4.1 Flash remains one of the most affordable frontier-grade models in the global AI landscape.
| Billing Category | Price per 1M Tokens | Conditions / Details |
|---|---|---|
| Input (Cache Miss - Off-Peak) | $0.15 / 1M tokens | Standard baseline rate (00:00-01:00, 04:00-06:00, 10:00-24:00 UTC) |
| Input (Cache Miss - Peak) | $0.30 / 1M tokens | Peak business hours (01:00-04:00 & 06:00-10:00 UTC) |
| Input (Cache Hit - Off-Peak) | $0.003 / 1M tokens | 🏆 98% discount for cached system prompts & repetitive context |
| Input (Cache Hit - Peak) | $0.006 / 1M tokens | 98% discount during peak hours |
| Output (Off-Peak) | $0.60 / 1M tokens | Standard text and code generation |
| Output (Peak) | $1.20 / 1M tokens | Peak generation rate |
| Batch API Input | $0.15 / 1M tokens | Asynchronous processing with guaranteed turnaround |
| Batch API Output | $0.60 / 1M tokens | Off-peak parity for async evaluation and batch workflows |
The $0.003 Prompt Caching Revolution
While competitors like OpenAI ($0.075/M cached) and Anthropic ($0.25/M cached) offer generous prompt caching discounts, DeepSeek has established a new market floor at $0.003 per million cached tokens.
Ingesting a 100,000-token company handbook or documentation base into 100 subsequent queries costs mere pennies once cached in memory.
Coding agents that constantly re-read git diffs, AST file trees, and system prompt instructions avoid paying repeated $0.15/M rates.
DeepSeek's Multi-Head Latent Attention (MLA) architecture compresses Key-Value (KV) caches directly in GPU VRAM, allowing the platform to pass infrastructure savings straight back to developers.
DeepSeek V4.1 Flash vs DeepSeek V4 Pro vs V3 vs Competitors
How does the new Flash model compare with previous generations and popular budget alternatives like GPT-4o mini and Claude Haiku?
| Model | Input / 1M | Cached In | Output / 1M | Context | Vision |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash Active (Latest) | $0.15 | $0.003 | $0.60 | 1.05M | Yes (Native 🏆) |
| DeepSeek V4 Pro Auto-Routed | $0.435 | $0.004 | $0.87 | 128k | Partial |
| DeepSeek V3 Active (Text) | $0.14 | $0.014 | $0.28 | 64k | No |
| DeepSeek R1 Active (Reasoning) | $0.55 | $0.14 | $2.19 | 64k | No |
| GPT-4o mini Active | $0.15 | $0.075 | $0.60 | 128k | Yes |
| Claude Haiku 3.5 Active | $0.80 | $0.08 | $4.00 | 200k | Yes |
For a complete ranking across all leading provider tiers, see our in-depth analysis of the Cheapest LLM APIs for Production in 2026 or explore technical specs in our 2026 AI Model Insights Hub.
DeepSeek V4.1 Benchmarks & Multimodal Visual Understanding
DeepSeek V4.1 Flash is not just a price play—it delivers generational architectural improvements over V4 Pro. By incorporating direct visual tokenizers and improved reinforcement learning with verifiable rewards, V4.1 Flash scores at or near frontier level across critical developer benchmarks:
Zero-shot Python synthesis matching proprietary models
Multi-step chain-of-thought mathematical reasoning
High-resolution diagram and document OCR parsing
How Much Do Vision Tokens Cost in DeepSeek Flash?
Images in DeepSeek V4.1 Flash are sliced into 384×384 pixel patches, each consuming approximately 256 input tokens. A standard 1080p screenshot decomposes into roughly 6 patches (~1,536 tokens), costing less than $0.00023 per image at baseline rates. For document OCR pipelines, this makes DeepSeek V4.1 Flash up to 15× more cost-effective than legacy proprietary vision APIs.
Real-World Cost Calculations (Worked Examples)
To illustrate the actual spend on an application running in production, let’s model three realistic monthly workloads using our LLM token cost calculator:
Scenario 1: High-Volume Customer Support Bot
50,000 conversations/month. Each request averages 800 input tokens (including system prompt) and 250 output tokens.
Scenario 2: Long-Context Document Analysis with Prompt Caching
A 50,000-token legal brief queried 200 times. With prompt caching, 199 calls hit the cache.
How to Call DeepSeek-V4.1-Flash in Code
DeepSeek maintains full drop-in compatibility with the official OpenAI SDK. Simply update your baseURL and select the deepseek-flash model identifier:
import OpenAI from "openai";
const openai = new OpenAI({
baseURL: "https://api.deepseek.com",
apiKey: process.env.DEEPSEEK_API_KEY,
});
const completion = await openai.chat.completions.create({
model: "deepseek-flash", // Official V4.1 Flash identifier
messages: [
{ role: "system", content: "You are an expert AI research assistant." },
{ role: "user", content: "Explain how to calculate token costs accurately." }
],
temperature: 0.3,
max_tokens: 4096,
});
console.log(completion.choices[0].message.content);Related 2026 AI Cost Guides & Benchmarks
Frequently Asked Questions (FAQ)
Q.How much does DeepSeek V4.1 Flash API cost?
DeepSeek V4.1 Flash costs $0.15 per 1 million input tokens ($0.30 during peak hours) and $0.60 per 1 million output tokens ($1.20 during peak hours). If your prompt hits the KV cache, input tokens cost just $0.003 per million tokens — an astonishing 98% discount.
Q.Is DeepSeek V4.1 Flash cheaper than DeepSeek V4 Pro?
Yes, significantly. DeepSeek V4.1 Flash reduces input costs from $0.435/M down to $0.15/M (a 65% savings) and output costs from $0.87/M down to $0.60/M (a 31% savings). Because V4.1 Flash outperforms V4 Pro in both speed and accuracy, DeepSeek is automatically routing legacy deepseek-v4-pro API calls to V4.1 Flash at the lower Flash price point.
Q.What is the context window limit of DeepSeek V4.1 Flash?
DeepSeek V4.1 Flash features a massive 1,048,576-token (~1.05 Million tokens) context window and supports up to 384,000 output tokens. This is an 8x expansion over DeepSeek V4 Pro's 128k limit and 16x larger than DeepSeek V3's 64k limit.
Q.Does DeepSeek V4.1 Flash support multimodal image understanding?
Yes. Unlike DeepSeek V3 and R1, DeepSeek V4.1 Flash comes with native multimodal vision capabilities, allowing you to pass high-resolution images, technical diagrams, charts, and scanned PDFs directly within the standard chat completions endpoint.
Q.What is the API model identifier for DeepSeek V4.1 Flash?
To call DeepSeek V4.1 Flash via official DeepSeek SDKs or OpenAI-compatible API clients, use the model name 'deepseek-flash' with baseURL 'https://api.deepseek.com'.
Q.When are DeepSeek API peak vs off-peak hours?
DeepSeek defines peak hours as Monday through Friday, 01:00 to 04:00 UTC and 06:00 to 10:00 UTC. Outside of these hours and throughout the entire weekend, all requests are billed at the discounted off-peak rate ($0.15/M input and $0.60/M output).
Simulate Your DeepSeek API Bill in Seconds
Compare DeepSeek V4.1 Flash against GPT-6 Astra, Claude Fable 5.1, Kimi K3, and 100+ AI models using our free, interactive token cost calculator with exact prompt caching discounts.