DeepSeek ships V4-Pro and raises its prices

Overview

Updated Aug 14

DeepSeek's flagship model left preview on August 13. Independent benchmark results are mixed: V4-Pro scored 53 on the Artificial Analysis Intelligence Index, trailing a mid-tier OpenAI GPT-5.6 model but roughly matching Zhipu AI's GLM-5.2, while beating Anthropic's Claude Fable 5 on cybersecurity tests.

The August 16 price increase is DeepSeek's clearest move yet from growth to profit. It comes as the company restarts a stalled $8 billion funding round at a $74 billion valuation, with signing expected by late August. Investors are betting DeepSeek can charge more without losing developers.

Why it matters

DeepSeek's low prices forced rivals to respond once. If it can raise prices and keep users, the whole AI cost war changes.

Questions about this story

Free account needed to ask — your question is kept and asked for you right after sign-up. Answers are public.

No questions yet — be the first to ask.

Key Indicators

1M tokens

Context window

How much text the model can read at once, roughly a long novel and then some.

~$3.96/M

Peak-hour price after increase

Cost per million tokens during peak hours once the August 16 increase takes effect.

~4x

Peak-hour cost jump

Rough increase in peak-hour costs from the model's preview pricing.

Artificial Analysis Intelligence Index score

V4-Pro's score on a widely tracked independent benchmark, about level with Zhipu AI's GLM-5.2 and behind a mid-tier GPT-5.6 model.

$74B

Target funding valuation

Valuation DeepSeek is seeking in its resumed $8 billion funding round, with signing expected by late August 2026.

Voices

Curated perspectives — historical figures and your fellow readers.

People Involved

Liang Wenfeng

Founder and chief executive, DeepSeek

Leading DeepSeek's push to monetize its flagship models

Organizations Involved

DeepSeek

AI company

Releasing its flagship model, raising prices, and closing a large funding round

A Hangzhou-based AI lab known for building competitive models cheaply and releasing them at low prices.

Timeline

July 2023 August 2026

7 events Latest: August 16th, 2026 · 3 months ago

Tap a bar to jump to that date

August 2026
Price increase takes effect
Latest Pricing

New V4 rates raise peak-hour costs roughly fourfold, to about $3.96 per million tokens, with steeper jumps for some token types.
August 16th, 2026
V4-Pro leaves preview
Product

DeepSeek makes V4-Pro (build 0813) generally available across its app, website and API, stressing agent tasks and a 1-million-token context window.
August 13th, 2026
Independent benchmarks show mixed results for V4-Pro
Benchmark

V4-Pro scores 53 on the Artificial Analysis Intelligence Index, trailing a mid-tier GPT-5.6 model but roughly matching Zhipu AI's GLM-5.2. It outperforms rivals, including Anthropic's Claude Fable 5, on cybersecurity benchmarks like CyberGym.
August 13th, 2026
DeepSeek resumes $8 billion funding round
Funding

DeepSeek restarts a fundraising round it had paused, targeting close to $8 billion at a roughly $74 billion valuation, with Monolith Management in talks to join. Signing is expected by late August.
August 6th, 2026
January 2025
US tech stocks plunge
Market

Nvidia falls about 17% and loses roughly $589 billion in market value as investors reassess AI spending.
January 27th, 2025
R1 model released
Product

DeepSeek releases its R1 reasoning model as open weights, matching top US systems at far lower reported cost.
January 20th, 2025
July 2023
DeepSeek founded
Origin

Liang Wenfeng starts DeepSeek, backed by his hedge fund High-Flyer, to build large language models.
July 1st, 2023

Scenarios

DeepSeek keeps the higher prices through year-end

Likely Resolves by End of 2026

Discussed by: Tech Startups, Unite.AI coverage of the pricing shift

DeepSeek holds its new, higher rates and does not walk them back. This would suggest the company believes V4-Pro's agent features are strong enough to keep developers paying, even after years of drawing them in with rock-bottom prices. It would mark a clear pivot from market share toward revenue.

A major US lab cuts flagship API prices in response

Possible Resolves by Feb 28, 2027

Discussed by: AI market analysts tracking OpenAI, Anthropic and Google pricing

OpenAI, Anthropic or Google lowers the price of a flagship model, a move competitors often make when a strong low-cost rival gains ground. DeepSeek's cheap models forced such responses before. A fresh cut would show V4-Pro is still shaping the wider market even as DeepSeek itself raises prices.

DeepSeek-V4-Pro ranks top five on a major public benchmark

Possible Resolves by End of 2026

Discussed by: Benchmark trackers such as LMArena and coding leaderboards

V4-Pro lands in the top five of a widely watched public leaderboard for agentic or coding tasks. That would back DeepSeek's claim that the model competes at the frontier and help justify the price increase to developers weighing cost against capability.

US government adds new restrictions on DeepSeek use

Uncertain Resolves by Aug 13, 2027

Discussed by: Policy reporting on US export controls and China AI

US federal action limits DeepSeek's use, such as a ban on government devices or new curbs tied to its Chinese ownership. Officials have raised data and security concerns about Chinese AI apps before. A fresh restriction would slow DeepSeek's US growth just as it tries to monetize.

DeepSeek closes its $8 billion round at a $74 billion valuation

Deadline passed (Aug 31, 2026)

Discussed by: Bloomberg and PYMNTS reporting on the funding talks

DeepSeek signs a deal for roughly $8 billion at close to a $74 billion valuation, with part of the money earmarked for a large data center in Inner Mongolia. Signing was expected by late August, according to Bloomberg's sourcing.

Historical Context

3 moments from history that rhyme with this story — and how they unfolded.

January 2025

DeepSeek R1 market shock (2025)

DeepSeek released its R1 reasoning model as open weights, claiming performance near top US systems at a reported training cost of a few million dollars. Investors feared AI could be built far more cheaply than assumed.

Then

On January 27, Nvidia fell about 17% and lost roughly $589 billion in market value in one day, though shares recovered part of the drop soon after.

Now

The episode made DeepSeek globally known and forced US labs and investors to defend their spending on expensive chips and training runs.

Why this matters now

It shows how much weight the market puts on DeepSeek's pricing. The V4-Pro increase reverses the very low-cost strategy that caused that shock.

May 2024

China's AI price war (2024)

Chinese tech firms including ByteDance, Alibaba and DeepSeek slashed model prices in a race for users. Some cut API costs by more than 90% within weeks.

Then

Prices for Chinese large language models fell sharply, pressuring margins across the industry.

Now

Low prices became a defining feature of China's AI market and a competitive threat to higher-priced US models.

Why this matters now

DeepSeek helped drive that price war. Raising V4-Pro prices now signals it may be stepping back from the discounting it once led.

2017-2019

Uber's shift from subsidies to profit (2017-2019)

Uber spent years offering rides below cost to win riders and drivers. After going public in 2019, it raised prices and cut incentives to chase profitability.

Then

Fares rose and some users complained, but Uber kept most of its customer base.

Now

The company eventually reached profitability, showing that a low-price disruptor can raise prices once it dominates a market.

Why this matters now

DeepSeek faces the same test: whether users who came for cheap AI will stay once prices climb.