Key Takeaways
- DeepSeek V4 Flash charges $0.14 per 1M input tokens, $0.28 per 1M output tokens, and $0.0028 per 1M cache-hit tokens on the official DeepSeek API.
- Cache-hit tokens cost a fraction of standard input rates, making repeated system prompts, RAG, and multi-turn chat significantly cheaper at scale.
- The model supports a 1M token context window and a 384,000 token maximum output length, with costs scaling linearly with usage.
- DeepSeek V4 Flash has no permanent free tier on the official API; new accounts receive trial credits only, and sustained access requires payment.
- OpenQiao currently offers DeepSeek V4 Flash at a promotional $0 rate across all provider slugs — the fastest route to free production-grade access at time of writing.
- Reasoning tokens bill at the output rate, but even multi-step reasoning calls producing thousands of tokens cost under one cent at V4 Flash rates.
- V4 Flash suits high-volume, cost-sensitive workloads; V4 Pro is the better choice when output quality is the binding constraint despite its higher rates.
DeepSeek V4 Flash API Pricing at a Glance
DeepSeek V4 Flash charges $0.14 per 1 million input tokens and $0.28 per 1 million output tokens. It also lists $0.0028 per 1 million cache-hit tokens and $0.14 per 1 million cache-miss tokens, as shown on the official DeepSeek pricing page .
| Pricing Dimension | Rate (per 1M tokens) |
|---|---|
| Input tokens | $0.14 |
| Output tokens | $0.28 |
| Cache-hit tokens | $0.0028 |
| Cache-miss tokens | $0.14 |
DeepSeek V4 Flash pricing reflects the official DeepSeek pricing page as fetched at the time of publication. DeepSeek updates its pricing schedule without advance notice, so verify the current schedule directly before committing to production workloads.
DeepSeek V4 Flash serves as the low-cost tier in DeepSeek's V4 lineup. Flash carries a materially lower output rate than V4 Pro, making it the default choice for high-throughput tasks where per-call cost dominates latency requirements. The cache-hit rate sits below the standard input rate, rewarding workloads that repeatedly send the same context prefix.
How Cache-Hit and Cache-Miss Pricing Lowers Your Cost
DeepSeek V4 Flash cache-hit tokens are billed at a fraction of the cache-miss rate. Any workload that reuses the same context window across multiple requests pays substantially less per token .
DeepSeek V4 Flash's prompt caching works by storing a previously processed token sequence in its KV cache. A cache-hit occurs when an incoming request prefix matches that stored sequence exactly. A cache-miss occurs when no match exists and the model processes the tokens from scratch, billing at the full input rate.
DeepSeek V4 Flash's cache pricing has the largest practical impact on 3 workload types:
- Repeated system prompts — a fixed instruction block sent with every request hits the cache on every call after the first
- Retrieval-augmented generation (RAG) — long document chunks prepended to each query reuse the same prefix across hundreds of requests
- Multi-turn chat — earlier turns in a conversation accumulate as a growing cached prefix, reducing the billable cost of each new message
DeepSeek V4 Flash supports 4 techniques to maximize cache hits. First, place stable content — system prompts, retrieved documents, few-shot examples — at the start of the prompt, before any variable user input. Second, keep the prefix byte-identical across requests; even a single character change invalidates the cache match.
Third, batch requests that share the same prefix within the same session window. Fourth, avoid shuffling retrieved chunks between calls, since order changes break prefix matching. Order matters more than it seems.
Context reuse through prompt caching is the highest-impact cost control available for DeepSeek V4 Flash at scale.
Context Window, Max Tokens & How They Affect Cost
DeepSeek V4 Flash supports a context window of 1M tokens and a maximum output length of 384,000 tokens .
DeepSeek V4 Flash prices input cost on a linear scale tied to prompt length. Every token sent in a request — system prompt, conversation history, and user message combined — is billed at the input rate. A prompt twice as long costs exactly twice as much in input charges.
DeepSeek V4 Flash prices output cost on a linear scale tied to the number of tokens the model generates. Shorter completions reduce output spend in direct proportion to their length.
DeepSeek V4 Flash includes a reasoning mode. Reasoning tokens — the internal chain-of-thought tokens the model generates before producing its final answer — are billed as output tokens . A token budget parameter controls the maximum number of reasoning tokens the model expends per request. Setting a tighter token budget caps the reasoning-token count and, by extension, caps the output-token charge for that reasoning work.
DeepSeek V4 Flash spend per call hinges on 3 practical levers: prompt length (input tokens), completion length (output tokens), and reasoning token budget (output tokens). Trimming any one of these 3 reduces the corresponding line item on the bill without changing the per-token rate.
Is There a Free DeepSeek V4 Flash API Tier?
There are 2 routes to free DeepSeek V4 Flash API access: OpenQiao's current promotional $0 rate and the official DeepSeek trial credits. They are not equivalent — one is active and production-capable right now, the other is a one-time allocation that expires.
OpenQiao currently offers DeepSeek V4 Flash at $0 per million tokens across all provider slugs, with standard rates struck through on the pricing page . This is a promotional rate active at time of writing — not a permanent free tier — but it applies to both input and output tokens with no published throughput cap, making it the most accessible route to zero-cost production calls available today. Verify current availability at openqiao.com/deepseek/deepseek-v4-flash-free before committing to production workloads, as promotional rates can change without notice.
The official DeepSeek platform grants new accounts a credit balance to test the API before any payment is required . Once those credits are exhausted, every subsequent call is billed at the standard per-token rates. DeepSeek V4 Flash API access on the official platform has no unlimited free tier — what exists is trial credits issued at account registration, not a permanent zero-cost route.
DeepSeek V4 Flash free access claims from aggregators originate from 2 sources: promotional trial credits bundled with new account sign-ups, and rate-limited free routes that cap throughput to discourage production use. Rate-limited free routes enforce strict request-per-minute ceilings, making them unsuitable for anything beyond exploratory testing.
DeepSeek V4 Flash's practical distinction matters: trial credits expire and carry a fixed token ceiling, while a true free tier would impose no expiry and no billing requirement regardless of volume. Developers who exhaust official trial credits without adding a payment method receive authentication errors on subsequent calls. OpenQiao's promotional $0 rate sidesteps this entirely — for as long as the promotion is active, there is no credit ceiling to exhaust on that platform.
DeepSeek V4 Flash Provider Comparison: Official API vs OpenRouter vs DeepInfra vs OpenQiao
OpenQiao delivers the most transparent, lock-in-free access to DeepSeek V4 Flash across all 4 providers compared here. The official DeepSeek API carries the lowest published base rates, but it binds developers to a single vendor with no fallback routing.
DeepSeek V4 Flash pricing appears below across sourced columns, with one qualitative hands-on judgment per provider.
| Provider | Input /1M tokens | Cache-Hit Rate | Free Tier | Hands-On Judgment | |---|---:|---:|---:|---|---| | Official DeepSeek API | $0.14 | $0.28 | $0.0028 | Trial credits only | Lowest base rate; single-vendor dependency with no automatic failover | | OpenRouter | $0.0896 | $0.1792 | Not published | Limited free credits | Adds a routing markup; broad model catalogue but pricing transparency varies by model | | DeepInfra | $0.09 | $0.18 | $0.018 cached | No | Competitive inference pricing; suited for high-throughput batch workloads with direct GPU access | | OpenQiao (via DeepSeek) | $0.14 $0 | $0.28 $0 | $0.0028 $0 | Promotional $0 rate active at time of writing | No lock-in; routes across DeepSeek, Baidu, and more — best for multi-model agent stacks that need provider resilience |
OpenQiao currently lists a 95.0% cache-hit rate for the DeepSeek provider slug , reflecting real-world prefix reuse across its routing layer. All rates for both the DeepSeek and Baidu provider slugs on OpenQiao show $0 / M tokens at time of writing (standard rates struck through on the OpenQiao pricing page) . A Baidu provider option is also available on OpenQiao (standard input $0.148/1M, standard output $0.296/1M, no published cache-read rate), though Baidu is marked as sunset . Verify current provider availability and promotional pricing at openqiao.com/deepseek/deepseek-v4-flash-free before committing to production workloads, as promotional rates can change without notice.
DeepSeek V4 Flash access through OpenQiao spans providers including DeepSeek, Baidu, and more (source). Vendor lock-in is eliminated by design. A single integration point covers the full provider roster (source).
DeepSeek V4 Flash access via the official API creates a hard dependency: authentication errors surface immediately when that single endpoint is unavailable, as confirmed in the previous section. OpenRouter and DeepInfra each introduce their own markup structures, which narrows the base-rate advantage of the official API at scale. OpenQiao resolves this by monitoring provider latency and availability in real time, routing each request to the fastest healthy endpoint without requiring a separate integration per vendor.
How We Compared V4 Flash Pricing Across Providers
DeepSeek V4 Flash API pricing comparisons here rest on published per-token rates from each provider's official pricing page. We pressure-tested them against real workload token math — a synthesis of public documentation plus hands-on API usage, not a lab measurement.
DeepSeek V4 Flash rate collection started at each provider's dashboard: the official DeepSeek platform, OpenRouter's model listing, DeepInfra's pricing index, and OpenQiao's provider listing. We recorded input, output, cache-hit, and cache-miss rates exactly as published at the time of writing.
DeepSeek V4 Flash workload mapping translated those rates into cost estimates across 3 representative task types: short-context classification calls, mid-length summarization jobs, and long-context document analysis. Each task type carries a distinct token distribution, so the same per-token rate produces a different effective cost depending on output-to-input ratio.
DeepSeek V4 Flash caching costs were factored in by running identical prompt prefixes across repeated calls, observing whether billing reflected the lower cache-hit rate. Providers that do not expose cache-hit billing were noted as opaque on that dimension.
DeepSeek V4 Flash qualitative judgment covered 2 dimensions beyond raw price. These are access friction (how many integration steps a developer completes before the first successful call) and pricing transparency (whether the dashboard surfaces a per-call cost breakdown in real time). These judgments reflect direct usage experience and carry no numeric measurement.
V4 Flash Cost Calculator: Worked Examples for Chat, Coding & Reasoning
A typical single-turn chat exchange with DeepSeek V4 Flash costs a fraction of a cent, making it one of the most cost-efficient models available for high-volume conversational workloads. The 3 worked examples below derive from the rates shown in the pricing table above; consult that table for the exact per-million-token figures.
There are 3 workload types to examine: chat, coding, and reasoning.
1. Chat Assistant
A standard chat turn sends roughly 200 input tokens and receives roughly 150 output tokens. At V4 Flash's input and output rates, the cost per turn lands well under $0.001. Running 10,000 such turns — a realistic daily volume for a small production chatbot — keeps the total bill in the low single-digit dollar range. That positions V4 Flash as genuinely inexpensive for conversational deployments at scale.
2. Coding Assistant
A coding session typically involves a long, stable system prompt — a repository context or a style guide — that repeats across every request. That repeated block qualifies for the cache-hit rate described in the pricing table, which is substantially lower than the standard input rate. A developer sending 2,000 input tokens per request, with 1,500 of those tokens served from cache, pays the cache-hit rate on the majority of each call. Over hundreds of daily completions, cache reuse compresses the effective per-request cost significantly relative to a naïve uncached calculation.
3. Reasoning Workload
Reasoning tasks generate thinking tokens in addition to visible output tokens. Both token streams bill at the output rate. A single reasoning call that produces a long chain-of-thought trace before delivering a final answer accumulates more output tokens than a direct-answer call of equivalent surface length. Even so, V4 Flash's output rate keeps the per-call cost competitive against models that charge a premium for reasoning capability. The decisive proof is that a reasoning call producing several thousand total output tokens still costs less than a cent at V4 Flash rates — a threshold that makes iterative, multi-step reasoning economically viable without per-query budgeting overhead.
V4 Flash vs V4 Pro: Pricing and Capability Tradeoff
DeepSeek V4 Flash is the right choice for high-volume, cost-sensitive workloads — classification, summarization, retrieval-augmented generation, and lightweight chat — where throughput volume makes per-token cost the dominant variable. DeepSeek V4 Pro is the right choice when task quality is the binding constraint: complex multi-step reasoning, nuanced code generation, and instruction-following tasks where a weaker output requires expensive human correction.
The price gap between the 2 tiers is substantial. V4 Flash carries a competitive input and output rate that keeps even reasoning calls under a cent at moderate token counts, as established in the previous section. V4 Pro's rates are meaningfully higher, reflecting its stronger parameter count and benchmark performance. A full breakdown of V4 Pro's input, output, and cache rates lives in the dedicated V4 Pro pricing article — this section does not duplicate those figures.
The decision rule is direct: choose V4 Flash when cost-per-query is the primary metric and output quality is acceptable for the task. Choose V4 Pro when output quality directly determines downstream value and the cost premium is recoverable through fewer retries or higher user retention.
For teams routing between both tiers dynamically, OpenQiao provides intelligent routing to official, authorized channels with no degraded model copies and verifiable, traceable quality per call. Its "Assured by OpenQiao" watermark signals safety at a glance, and its built-in compensation mechanism turns shortfalls into a data flywheel rather than silent failures. Developers shipping production AI apps or agents — and teams needing unified access to DeepSeek, OpenAI, Anthropic, Google, and more through one API with no vendor lock-in — are the primary fit. Explore the routing layer at openqiao.com.
How to Access and Call the DeepSeek V4 Flash API
DeepSeek V4 Flash is available through an OpenAI-compatible REST endpoint. This means any application already using the OpenAI SDK requires only 2 changes to switch: swap the base URL to DeepSeek's API host (source) and set the model name to deepseek-v4-flash . No SDK replacement, no rewritten request logic.
DeepSeek V4 Flash uses the base URL https://api.deepseek.com for DeepSeek's official API, and the model name string for V4 Flash is deepseek-v4-flash as confirmed in DeepSeek's official documentation . Authentication uses a standard Bearer token passed in the Authorization header, identical to the OpenAI pattern.
OpenQiao provides access to DeepSeek V4 Flash through the same OpenAI-compatible interface, with no vendor lock-in. Supported providers on OpenQiao include DeepSeek, Baidu, and more. A single OpenQiao API key routes calls to any of these providers. Switching away from DeepSeek — or routing between providers dynamically — requires no code change beyond the model name.
For a complete walkthrough covering authentication, request structure, and response parsing, see the dedicated DeepSeek V4 Flash how-to guide on openqiao.com.
Rate Limits and Reasoning-Token Billing
DeepSeek V4 Flash API enforces concurrency-based rate limits that scale with account credit balance . Reasoning tokens — the internal "thinking" tokens the model generates before producing a visible response — are billed at the standard output token rate .
DeepSeek V4 Flash API rate limits operate on a concurrency model: the default account-level concurrency limit for deepseek-v4-flash is 2,500 simultaneous requests . Accounts with higher prepaid credit balances can request elevated limits through a capacity expansion request, without requiring a manual tier upgrade.
DeepSeek V4 Flash API reasoning-token billing has a direct impact on cost predictability for tasks that trigger extended chain-of-thought. A single reasoning-heavy prompt generates a larger total output token count than its visible text alone suggests. Callers who set a max_tokens ceiling on the combined output control the upper bound of that cost exposure. Without that ceiling, a complex reasoning task expands the billable output token count significantly beyond the length of the final answer text.
Frequently Asked Questions
How much does the DeepSeek V4 Flash API cost per million input and output tokens?
DeepSeek V4 Flash API pricing is $0.14 per million input tokens and $0.28 per million output tokens at the official DeepSeek platform . Cache-hit input tokens are billed at $0.0028 per million tokens, reducing costs for repeated prompt prefixes. OpenQiao currently lists all rates at a promotional $0 / M tokens — verify current availability before relying on it for production workloads.
Can I use DeepSeek V4 Flash for free, or is it only trial credits and rate-limited routes?
There are 2 practical routes to free access right now. OpenQiao currently lists a promotional $0 rate for all DeepSeek V4 Flash provider slugs at time of writing — this is the most accessible zero-cost option available today, with no published throughput cap. Verify current availability at openqiao.com/deepseek/deepseek-v4-flash-free before relying on it for production workloads, as promotional rates can change without notice. The official DeepSeek platform is not permanently free; new accounts receive a limited trial credit allocation that expires after a fixed period. Rate-limited free routes exist on some third-party aggregators, but those routes enforce strict request-per-minute caps that make them unsuitable for production workloads.
Which providers offer DeepSeek V4 Flash, and which is the cheapest?
DeepSeek V4 Flash is available through the official DeepSeek API, OpenRouter, DeepInfra, and OpenQiao. Per-token rates differ across these 4 providers; the comparison table in the provider section above shows the exact figures with sources for each. OpenQiao currently shows a promotional $0 rate across all provider slugs , making it the lowest-cost option at time of writing.
How much cheaper is V4 Flash than V4 Pro, and when should I pick each?
DeepSeek V4 Flash input and output token prices are substantially lower than V4 Pro rates . This makes Flash the correct choice for high-volume, latency-sensitive tasks where the quality ceiling of Pro is not required.