DeepSeek API Pricing (July 2026): V4 Flash $0.14/M · V4 Pro $0.435/M

Verified pricing data

DeepSeek API prices as of July 2026

Source: https://api-docs.deepseek.com/quick_start/pricing/

As of

July 31, 2026

Update cadence

Checked daily; published only after source verification.

Source policy

Official provider pricing pages or APIs only.

Machine-readable data

/api/pricing.json

Tracked sources

https://api-docs.deepseek.com/quick_start/pricing/

Estimate monthly costs with these prices

Quick pricing answers

Cheapest input

DeepSeek V4 Flash

$0.14/M input · $0.28/M output

Cheapest output

DeepSeek V4 Flash

$0.28/M output

Best cached-input price

DeepSeek V4 Flash

$0.0028/M cached input

Widest context

DeepSeek V4 Flash

1,000,000 tokens context

These answers are generated from 2 verified DeepSeek models as of July 2026. Use the calculator for workload-specific totals because output tokens and cache hit rate can change the cheapest choice.

Model	Input / 1M	Cached input / 1M	Output / 1M	Context	Max output
DeepSeek V4 Flash	$0.14	$0.0028	$0.28	1,000,000	384,000
DeepSeek V4 Pro	$0.435	$0.0036	$0.87	1,000,000	384,000

Official DeepSeek-V4-Flash-0731 public-beta pricing. The callable model ID remains deepseek-v4-flash. It supports thinking and non-thinking modes, the Responses API, a 1M-token context window, and up to 384K output tokens. DeepSeek has announced future 2x pricing during peak hours, but no effective date is set, so this row uses the current regular rates.

Official DeepSeek V4 Pro pricing. It supports thinking and non-thinking modes, a 1M-token context window, and up to 384K output tokens. Responses API support is planned for early August 2026. DeepSeek has announced future 2x pricing during peak hours, but no effective date is set, so this row uses the current regular rates.

Pricing questions

What is the cheapest DeepSeek API model in July 2026?

DeepSeek V4 Flash has the lowest verified DeepSeek input-token price at $0.14/M input tokens.

Which DeepSeek model is cheapest for generated output?

DeepSeek V4 Flash has the lowest verified DeepSeek output-token price at $0.28/M output tokens.

Where does TLDL get DeepSeek pricing data?

TLDL uses official DeepSeek pricing sources and records source URLs plus verification dates in the shared pricing dataset.

DeepSeek pricing changelog

2026-07-31

Verified unchanged V4 Flash and V4 Pro token prices after the DeepSeek-V4-Flash-0731 public-beta update. DeepSeek also announced a future 2x peak-hours policy, but has not set its effective date.

https://api-docs.deepseek.com/quick_start/pricing/

2026-07-04

Verified DeepSeek V4 Flash and V4 Pro pricing from the official DeepSeek API pricing page.

https://api-docs.deepseek.com/quick_start/pricing

DeepSeek's public API exposes V4 Flash and V4 Pro through OpenAI-compatible and Anthropic-compatible endpoints. On July 31, 2026, DeepSeek moved the updated DeepSeek-V4-Flash-0731 API into public beta without changing its callable model ID or current token prices.

The live price table above is generated from TLDL's shared pricing data. The same verified rows power the LLM API pricing comparison, public pricing API, and LLM cost calculator.

DeepSeek API price at a glance

Prices are in US dollars per 1 million tokens.

These are the current regular rates shown in DeepSeek's official pricing table. DeepSeek has announced that it will introduce 2× pricing during peak hours—09:00–12:00 and 14:00–18:00 Beijing time (UTC+8)—for every billing item. The company has not announced an effective date, so TLDL does not apply the multiplier yet.

What changed on July 31

The deepseek-v4-flash model ID now serves DeepSeek-V4-Flash-0731 in public beta. The calling method and model name stay the same. DeepSeek says the update retains the preview model's architecture and size while improving its post-training, particularly for agent work.

V4 Flash now natively supports the Responses API and is adapted for Codex. V4 Pro and the DeepSeek app and web models were not changed by this release. DeepSeek says the official V4 Pro release will follow soon.

Limits and API support

Both models support thinking and non-thinking modes, with thinking enabled by default. They also support JSON output, tool calls, and chat-prefix completion. DeepSeek expects V4 Pro Responses API support in early August 2026.

When DeepSeek makes sense

DeepSeek is strongest when cost is a hard constraint and your application can tolerate some provider-specific behavior. It is especially useful for high-volume classification, routing, extraction, and batch analysis jobs where a small per-token difference becomes material.

For agent workflows, test tool-calling and latency before routing all traffic to DeepSeek. A cheap model that adds retries or manual review can cost more than it saves.

Migration notes

DeepSeek exposes an OpenAI-compatible API, so many clients only need a base URL and model-name change. Use deepseek-v4-flash for the latest Flash release, keep the model identifier configurable, and log token usage before and after migration.

If your app depends on reasoning behavior, verify whether you want thinking mode enabled for every request. Reasoning tokens are billed as output, so long reasoning traces can become the largest part of a request's cost. Budget for the announced peak-hours multiplier before it takes effect, even though current bills still use the regular rates.

Related

• LLM API Pricing Comparison
• LLM Cost Calculator
• OpenAI API Pricing
• Claude API Pricing