conv.

All stories
AIRunning 9d ↳ DeepSeek's Pricing Strategy Shift

DeepSeek Warns of Steep Price Hike Days After Cut-Rate V4 Flash Launch

The Chinese lab's ultra-cheap V4 Flash model intensified the AI price war, then DeepSeek told users to expect a 'significant' price increase.

15 items625 engagement336 comments Hacker NewsTechmemePress
DeepSeek Warns of Steep Price Hike Days After Cut-Rate V4 Flash Launch
Mashable

Conversation activity · last 10 days peak 6/2h

Peak 6 items in one 2h at Jul 31, 9 AM; 15 items over 10 days Jul 31, 3 AM — 1 itemJul 31, 5 AM — 2 itemsJul 31, 7 AM — no itemsJul 31, 9 AM — 6 itemsJul 31, 11 AM — no itemsJul 31, 1 PM — no itemsJul 31, 3 PM — no itemsJul 31, 5 PM — no itemsJul 31, 7 PM — 1 itemJul 31, 9 PM — no itemsJul 31, 11 PM — no itemsAug 1, 1 AM — no itemsAug 1, 3 AM — no itemsAug 1, 5 AM — no itemsAug 1, 7 AM — 1 itemAug 1, 9 AM — no itemsAug 1, 11 AM — no itemsAug 1, 1 PM — no itemsAug 1, 3 PM — no itemsAug 1, 5 PM — no itemsAug 1, 7 PM — no itemsAug 1, 9 PM — no itemsAug 1, 11 PM — no itemsAug 2, 1 AM — no itemsAug 2, 3 AM — no itemsAug 2, 5 AM — no itemsAug 2, 7 AM — no itemsAug 2, 9 AM — no itemsAug 2, 11 AM — no itemsAug 2, 1 PM — no itemsAug 2, 3 PM — no itemsAug 2, 5 PM — no itemsAug 2, 7 PM — no itemsAug 2, 9 PM — no itemsAug 2, 11 PM — no itemsAug 3, 1 AM — no itemsAug 3, 3 AM — no itemsAug 3, 5 AM — no itemsAug 3, 7 AM — no itemsAug 3, 9 AM — 1 itemAug 3, 11 AM — no itemsAug 3, 1 PM — no itemsAug 3, 3 PM — no itemsAug 3, 5 PM — no itemsAug 3, 7 PM — no itemsAug 3, 9 PM — no itemsAug 3, 11 PM — no itemsAug 4, 1 AM — no itemsAug 4, 3 AM — no itemsAug 4, 5 AM — no itemsAug 4, 7 AM — no itemsAug 4, 9 AM — no itemsAug 4, 11 AM — no itemsAug 4, 1 PM — no itemsAug 4, 3 PM — no itemsAug 4, 5 PM — no itemsAug 4, 7 PM — no itemsAug 4, 9 PM — no itemsAug 4, 11 PM — no itemsAug 5, 1 AM — no itemsAug 5, 3 AM — no itemsAug 5, 5 AM — no itemsAug 5, 7 AM — no itemsAug 5, 9 AM — no itemsAug 5, 11 AM — no itemsAug 5, 1 PM — no itemsAug 5, 3 PM — no itemsAug 5, 5 PM — no itemsAug 5, 7 PM — no itemsAug 5, 9 PM — no itemsAug 5, 11 PM — 1 itemAug 6, 1 AM — 1 itemAug 6, 3 AM — no itemsAug 6, 5 AM — no itemsAug 6, 7 AM — no itemsAug 6, 9 AM — no itemsAug 6, 11 AM — no itemsAug 6, 1 PM — no itemsAug 6, 3 PM — no itemsAug 6, 5 PM — no itemsAug 6, 7 PM — no itemsAug 6, 9 PM — no itemsAug 6, 11 PM — no itemsAug 7, 1 AM — no itemsAug 7, 3 AM — no itemsAug 7, 5 AM — no itemsAug 7, 7 AM — no itemsAug 7, 9 AM — no itemsAug 7, 11 AM — no itemsAug 7, 1 PM — no itemsAug 7, 3 PM — no itemsAug 7, 5 PM — no itemsAug 7, 7 PM — no itemsAug 7, 9 PM — no itemsAug 7, 11 PM — no itemsAug 8, 1 AM — no itemsAug 8, 3 AM — no itemsAug 8, 5 AM — no itemsAug 8, 7 AM — no itemsAug 8, 9 AM — no itemsAug 8, 11 AM — no itemsAug 8, 1 PM — no itemsAug 8, 3 PM — no itemsAug 8, 5 PM — no itemsAug 8, 7 PM — no itemsAug 8, 9 PM — no itemsAug 8, 11 PM — no itemsAug 9, 1 AM — no itemsAug 9, 3 AM — no itemsAug 9, 5 AM — no itemsAug 9, 7 AM — no itemsAug 9, 9 AM — no itemsAug 9, 11 AM — no itemsAug 9, 1 PM — 1 itemAug 9, 3 PM — no itemsAug 9, 5 PM — no itemsAug 9, 7 PM — no items 6 items · 9 AM
Aug 1Aug 2Aug 3Aug 4Aug 5Aug 6Aug 7Aug 8Aug 9

Summary, timeline and people extracted by Claude from 15 items across 3 sources · 5h ago. Quotes are verbatim.

DeepSeek released V4 Flash 0731 on July 31, an open-weights model that benchmarked near the top of the field while costing a fraction of Western rivals, accelerating an industry-wide 'race to zero' on AI pricing and reportedly pushing OpenAI to cut its own prices. Less than a week later, DeepSeek notified users it plans substantial price increases across its services without specifying the size of the hike, a reversal that has drawn speculation about DeepSeek's strategy and the sustainability of its pricing model as it builds a large data center in Inner Mongolia.

  • DeepSeek's V4 Flash 0731, released July 31, ranked #3 in intelligence and #1 in cache-hit price among 101 benchmarked models while costing $0.14/$0.28 per million input/output tokens.
  • The release intensified an AI 'price war,' with commentators linking it to OpenAI's preemptive price cuts and broader commoditization of AI intelligence.
  • On August 6, DeepSeek notified users of a coming 'significant' price increase without specifying the amount, reported first by Bloomberg.
  • Observers are divided on motive: some see the hike as inevitable given DeepSeek's costly Inner Mongolia data center plans and thin margins, others as a marketing tactic to boost usage before prices rise.

How it unfolded

  1. Report Mashable details price context

    Mashable summarized the Bloomberg report, noting DeepSeek currently charges under a dollar per million tokens versus Anthropic's Fable 5 at $10/$50 per million, and linked the hike to costs of a planned Inner Mongolia data center.

  2. Event DeepSeek announces coming price hike

    DeepSeek sent users a notice, reported first by Bloomberg, saying it plans a substantial increase to its API prices without giving specifics.

  3. Reaction HN debates motive behind the hike notice

    A commenter suggested the announcement is a marketing tactic to spur usage before the increase takes effect, comparing it to pricing dilemmas faced by Kimi and GLM.

    “This notice from DeepSeek might actually be a brilliant marketing move.”

    alexwwang · Hacker News ↗
  4. Report Semafor: China price war intensifies

    Semafor noted DeepSeek's roughly 50% cut in token costs as a bid for market share amid intensifying competition among Chinese AI labs.

  5. Analysis Axios: 'race to zero' accelerates

    Axios reported the cheap new coding model as further evidence that AI intelligence is rapidly commoditizing despite huge infrastructure spending.

  6. Report Analysts call it best value-per-intelligence model

    Simon Willison noted the 304B-parameter (per Hugging Face listing) model punches above its weight, ranking ahead of larger models like MiniMax M3 on Artificial Analysis's intelligence-vs-cost metric.

    “This may currently be the best value-per-intelligence model out there.”

    Simon Willison · Press ↗
  7. Event DeepSeek releases V4 Flash 0731

    DeepSeek published an open-weights reasoning model, 284B total/13B active parameters, with a 1M-token context window, priced at $0.14 per 1M input and $0.28 per 1M output tokens.

  8. Reaction HN speculates on competitive pressure

    Commenters linked the release to OpenAI's aggressive price cuts the prior day and speculated a stronger DeepSeek V4 Pro could follow.

    “This seems to me like this is probably at least a large part of what OpenAI was up to yesterday with their aggressive price cutting; trying to get out in front of this.”

    cmrdporcupine · Hacker News ↗

What people are saying verbatim

“My take: this notice from DeepSeek might actually be a brilliant marketing move.”

alexwwang, HN commenter · Hacker News ↗ · Aug 5

“This may currently be the best value-per-intelligence model out there.”

Simon Willison, AI commentator/blogger · simonwillison.net ↗ · Jul 30

“This seems to me like this is probably at least a large part of what OpenAI was up to yesterday with their aggressive price cutting; trying to get out in front of this.”

cmrdporcupine, HN commenter · Hacker News ↗ · Jul 30

“So GLM 5.2/Gemini 3.6 level intelligence for $0.28/m output. And their updated Pro model coming soon....”

scosman, HN commenter · Hacker News ↗ · Jul 30

“DeepSeek, famous for its cheap AI services, is about to become less cheap.”

Mashable, Tech publication · Mashable ↗ · Aug 8

“A 50% cut in token costs suggests DeepSeek is competing for market share.”

Semafor, News publication · Semafor ↗ · Aug 2

Voices from the web unedited

  • This seems to me like this is probably at least a large part of what OpenAI was up to yesterday with their aggressive price cutting; trying to get out in front of this.If the full non-flash model follows up with the expected improvements, and at the price point they've been keeping, it puts the frontier labs in a tough position and it feels to me…

    cmrdporcupineHacker News9d agoview on Hacker News ↗

  • > For the Code Agent tasks among the public benchmarks above, DeepSeek-V4-Flash-0731 is evaluated with the minimal mode of DeepSeek Harness (to be released) as the agent frameworkSo, are they planning to announce an optimized coding agent harness as well ? DSv4 flash is a fantastic model, and my daily driver. With reasonix or pi, I can code all…

    bwfan123Hacker News9d agoview on Hacker News ↗

  • I have updated OpenAI's chart[1] from yesterday to include one more datapoint: DeepSeek V4 Flash 0731. It's on the frontier.

    pmxiHacker News9d agoview on Hacker News ↗

  • The really interesting thing about this is how big of a jump was achieved with just extra fine-tuning here. No structural changes to the model, just more data, compute and time. It makes me pretty excited for the future of small models - DS v4 flash is a relatively small model when compared to the class it's competing with, so likely similar gains…

    kamranjonHacker News9d agoview on Hacker News ↗

  • Somewhat relatedly, how do the economics for Huggingface work? They must be hosting petabytes of models and datasets by now. I have downloaded quite a few “just in case”, only to replace them with the later iteration months later.Does the file hosting actually cost peanuts when you do it yourself and the cloud has shattered my understanding of…

    0cf8612b2e1eHacker News9d agoview on Hacker News ↗

  • So GLM 5.2/Gemini 3.6 level intelligence for $0.28/m output. And their updated Pro model coming soon....Plus a size you can genuinely run at home: Unsloth lossless Q8 at 162GB.

    scosmanHacker News9d agoview on Hacker News ↗

  • If deepseek v4 flash is beating DeepSeek V4 Pro, can we expect new V4 Pro which is on par with Opus 5 in couple weeks (even better if it beats Opus)?

    throwaw12Hacker News9d agoview on Hacker News ↗