DeepSeek
Chinese AI model that demonstrated competitive performance with Western AI models.
What else do they all swear by?
Riley Brown, Greg Isenberg, Jason Lemkin and 2 others use or recommend DeepSeek. See what else they have in common.
See what they all swear by Or pick your own peopleWhat creators think
Recently discussed in
How DeepSeek compares
18 creators weighed in
Kimi
Direct rival
Moonshot AI's large language model
DeepSeek costs $1 per million tokens; Kimi K3 costs $15. DeepSeek runs on a laptop; Kimi K3 needs massive GPU infrastructure.
G
R
+3
7 creators
G
R
+3
7 creators
DeepSeek costs $1 per million tokens; Kimi K3 costs $15. DeepSeek runs on a laptop; Kimi K3 needs massive GPU infrastructure.
G
R
+3
G
R
+3
"When was that last big hit to their valuations? When you had the DeepSeek moment between DeepSeek and Kimmy K3 — you see like these real assaults on these frontier models being able to justify their high price."
Cited alongside 'Kimmy K3' as models that challenged the pricing and valuations of frontier models like Anthropic and OpenAI
"If you look at the three big Chinese open source models — Deepseek, Kimi, Qwen — they're kind of all evolving in very different ways."
Grouped with Kimi and Qwen as the three major Chinese open-source models
"We keep finding ways to bend that curve. So, it's a little bit less bad than log linear, but it's still somewhere between a power law — as we see with DeepSeek Flash that just came out, as we see with Kimi K3. We keep finding ways to ben..."
Mentioned alongside Kimi K3 as examples of efficiency improvements
"DeepSeek V4 Flash is similar — from last week. That one's an even bigger deal because it was just pretrained and it fits on a MacBook."
Compared to Kimmy K3 as similarly capable open-weight models
"Let's take some of the DeepSeek models — you can run on your PC or your Mac or on a laptop. Conversely, Kimi K3 is a 2.8 trillion parameter model; it's a huge honking thing and you need myriads of GPUs just to run it."
Contrasted with Kimi K3 (2.8 trillion parameters) which requires massive GPU infrastructure
OpenAI
Direct rival
AI research and model company
DeepSeek is far cheaper. Ed says 96% cheaper than OpenAI; Naval uses it for most tasks because of the low cost. John finds OpenAI more reliable.
M
+2
6 creators
M
+2
6 creators
DeepSeek is far cheaper. Ed says 96% cheaper than OpenAI; Naval uses it for most tasks because of the low cost. John finds OpenAI more reliable.
M
+2
M
+2
"We took the cheapest DeepSeek model and made it beat OpenAI's last frontier model using this harness, and the cost is 50 times lower."
Beat OpenAI's previous frontier model
"So one day I'll plug in you know an OpenAI ChatGPT 4o open source and the next day I'll plug in DeepSeek"
Used alongside OpenAI models in a model-agnostic setup
"On the record day that Roush is talking about, he said spending on models from three Chinese labs — Moonshot, Deepseek, and Z.AI — added up to more than the spending on OpenAI."
Grouped with Moonshot and Z.AI as Chinese open weight models whose combined spend exceeded OpenAI on Vercel's record day
"Yes, I could just download the weights for DeepSeek, but I'm not really comfortable with that. I don't really want to deal with that. I don't really want to deal with the scaling and uptime of that."
Compared unfavorably to frontier labs like OpenAI, Anthropic, and Google in terms of reliability, SLA, and enterprise comfort.
"That allows them to build these incredibly cheap models like DeepSeek and Moonshot and then put them back as open-source models in the US."
Compared to OpenAI and Anthropic models as cheaper open-source alternatives; mentioned alongside Moonshot
Claude
Direct rival
Anthropic's AI assistant
Riley says DeepSeek is 20 to 105x cheaper than Claude. Scott notes DeepSeek Pro costs $3 vs $12.50 for Claude Opus.
S
4 creators
S
4 creators
Riley says DeepSeek is 20 to 105x cheaper than Claude. Scott notes DeepSeek Pro costs $3 vs $12.50 for Claude Opus.
S
S
"DeepSeek V4 Flash is probably the most interesting because for certain tasks it is 105 times cheaper than Claude. I highly recommend trying these models. This model is 20 to 100 times cheaper depending on the task."
Compared against Claude (105x cheaper); price increase warning from DeepSeek noted; OpenCode claims they can reproduce current pricing on rented GPUs
"DeepSeek V4 Pro costs about $3 versus $12.50 for Opus 4.5. Many of the strongest open-weight models now come from Chinese developers, including DeepSeek and Moonshot."
Compared to Claude Opus 4.5 on cost
"DeepSeek released a new pro model last night and talked like it was almost as good as the top expensive models like Fable, and it was only a tiny fraction of the price. However, many people started testing this, and they came back and sa..."
Compared to Claude Fable (claimed to be near-equivalent); actually found to be only slightly better than DeepSeek V4 Flash
"Please, somebody ask producer Claude or producer Grok or producer Deepseek, is anybody protesting these solar farms?"
Mentioned alongside Claude and Grok as AI research tools
"Get ready for Deep Seek 4, 20x over current US company LLM, even Opus 4.6 and Codex."
Compared to Claude Opus 4.6 and Codex
ChatGPT
Direct rival
OpenAI's conversational AI chatbot
DeepSeek is 10 to 50 times cheaper per token than GPT, per Peter and Ed.
J
E
4 creators
J
E
4 creators
DeepSeek is 10 to 50 times cheaper per token than GPT, per Peter and Ed.
"DeepSeek Flash is a great example and I think there'll be quite a few more where it's ultra low cost per token. It's also low cost per task, which is a really important metric. This new class of flash models where they're good enough for..."
Compared to the GPT-4o mini model as another example of a very price-effective model that enables widespread team adoption; positioned as leading the open model efficiency charge
"The elephant in the room is clearly the Chinese models. It is clearly DeepSeek, Kimi, all of these Chinese models, which are so dramatically cheaper than the US models that it's hard to actually fathom."
Compared to OpenAI GPT-5 ($45/million tokens) and Claude ($50/million tokens)
"OpenRouter allows us to only use one key, so we can get access to GPT 5.5, Claude Opus, DeepSeek V4, and in this case the most important one, GLM 5.2, and then thousands of other models."
Listed alongside GPT 5.5, Claude Opus, and GLM 5.2 as models accessible via OpenRouter
"You can run the DeepSeek V4 Pro model in your AI agents for 23 times less than the anthropic opus 4.8."
Compared to Anthropic Opus 4.8 and OpenAI GPT 5.5 for cost
"DeepSeek V4, number three in the world versus US models. A trillion parameters, 37 billion active parameters per token. It's 10 to 50 times cheaper than GPT 5.4 and Opus 4.6"
Compared to GPT and Claude models for cost efficiency
Anthropic
Direct rival
AI safety company
DeepSeek is building its own chips and hiring fast. Both have taken large outside funding rounds.
H
2 creators
H
2 creators
DeepSeek is building its own chips and hiring fast. Both have taken large outside funding rounds.
"DeepSeek is developing its own chip. And today DeepSeek has announced that they are starting to build their own chips."
Compared to Anthropic also exploring custom chips with Samsung
"DeepSeek has just announced plans to double its headcount in some departments and is preparing to take outside investment for the first time ever."
Compared to Anthropic and OpenAI in terms of funding round sizes.
Which version
Each new one replaces the last.
38 creators didn't say which. Their quotes are labeled with when they said it, so you can tell which was out at the time.
What creators say
"We took the cheapest DeepSeek model and made it beat OpenAI's last frontier model using this harness, and the cost is 50 times lower."
More from Emad Mostaque →+5 more mentions from Emad
"We released a harness called Zenith that takes DeepSeek v4.1 above Astra and it's getting up to Astra levels with open source models, but it kind o..."
"Deep Seek V4 Pro scores 83.3 on CyberSec whereas Methos scored 83.2. The capability is open source."
"When I'm in China for two weeks, it's DeepSeek and Gemini. Those are my friends. DeepSeek is intentionally crippled in China. It's not as good as it is here. It can't search the web and it is trained on different data as near as I can tell."
More from Jason Lemkin →"So one day I'll plug in you know an OpenAI ChatGPT 4o open source and the next day I'll plug in DeepSeek"
More from Mo Gawdat →
"DeepSeek V4 Flash is probably the most interesting because for certain tasks it is 105 times cheaper than Claude. I highly recommend trying these models. This model is 20 to 100 times cheaper depending on the task."
More from Riley Brown →+2 more mentions from Riley
"OpenRouter allows us to only use one key, so we can get access to GPT 5.5, Claude Opus, DeepSeek V4, and in this case the most important one, GLM 5..."
"You can run the DeepSeek V4 Pro model in your AI agents for 23 times less than the anthropic opus 4.8."
"DeepSeek. This is quite good at hard thinking and coding problems. But heads up, the reasoning models take 10 to 30 seconds to think before they answer. And that's normal."
More from Greg Isenberg →"DeepSeek V4 Pro costs about $3 versus $12.50 for Opus 4.5. Many of the strongest open-weight models now come from Chinese developers, including DeepSeek and Moonshot."
More from Scott Galloway →"It wasn't until DeepSeek V3 that it started being treated as a serious contender. And part of what made DeepSeek V3 so successful was that it was open."
More from Misha Laskin →
"So if China is actually using distillation attacks to make things like DeepSeek so powerful, that just got a lot harder."
More from Tom Bilyeu →+6 more mentions from Tom
"On the record day that Roush is talking about, he said spending on models from three Chinese labs — Moonshot, Deepseek, and Z.AI — added up to more..."
"When was that last big hit to their valuations? When you had the DeepSeek moment between DeepSeek and Kimmy K3 — you see like these real assaults o..."
"DeepSeek definitely put out a paper describing some of their optimizations. And then suddenly GPTs get faster like a month later."
More from Simon Willison →"September 9th, DeepSeek 4.1 Flash gets released. 20 cents for a million tokens output if they're hosting it. Key value cache from V1 of DeepSeek — remember what a big deal DeepSeek was when it came out — the key value cache then was like 390,000 b..."
More from David Friedberg →"Deep Seek beginning of 2025 with Deepseek V3 made a lot of waves because they were able to get a massive decrease in KV cache size with multi-head latent attention."
More from Thomas Sohmers →"It started really last year I would say with DeepSeek to really catch on, and this is the year of open source — it's crazy, this year has been for us like riding a tsunami wave of open models."
More from Thomas Wolf →
"There was an interesting post that hit the timeline. Was that actually from a DeepSeek employee? A DeepSeek employee basically was like, I don't like Dario. He compared giving Dario AGI to like giving Hitler the bomb."
More from John Coogan →+6 more mentions from John
"Yes, I could just download the weights for DeepSeek, but I'm not really comfortable with that. I don't really want to deal with that. I don't reall..."
"All of this is driven by the actual progress in open source. We wouldn't be having this discussion if Moonshot, ZAI, and many others, DeepSeek, wer..."
"The first is on DeepSeek V4.1 Flash showing how memory efficiency is increasing, reducing dependency on high bandwidth memory HBM."
More from Peter Diamandis →+1 more mention from Peter
"DeepSeek V4, number three in the world versus US models. A trillion parameters, 37 billion active parameters per token. It's 10 to 50 times cheaper..."
"The new DeepSeek model that came out last night — the reason that's a crazy dominating model is because they spent all the time on amazing data. That's it."
More from Om Buddhdev →"Back to Liang Wenfeng, founder of DeepSeek, his leaked transcript. It felt like there were a lot of interesting, thought-provoking ideas in there. He's the one player in this ecosystem who's prioritizing keeping his independence and research direc..."
More from Jordan Schneider →"If you look at the three big Chinese open source models — Deepseek, Kimi, Qwen — they're kind of all evolving in very different ways."
More from Gavin Baker →+1 more mention from Gavin
"Somebody told me it only took 160,000 reasoning traces from o1 and o3 to get the original DeepSeek. They're very clever at industrial scale distill..."
"Hackers are budget constrained. They can't just run Fable 5 all day and do all the things you're saying, but they can now afford to use Deepseek. And as those models get not just cheaper, but much more powerful, we're going to see this just contin..."
More from This Week in Startups →+2 more mentions from This
"Please, somebody ask producer Claude or producer Grok or producer Deepseek, is anybody protesting these solar farms?"
"The models from Nvidia's Neatron family, the models from Google's Gemma family, aren't sufficient to replace what we have from Deepseek, from Zhipu..."
"Deep Seek certainly publishes really interesting work to compress the KV cache further and further. They're making good progress."
More from Neil Movva →"I was talking to friends at DeepSeek a little bit after they came out, and before that nobody knew who they were. They were not on the radar, they were not funded by the government. Nobody told them to open source, but the CEO of the company is ve..."
More from Alvin Wang Graylin →"We keep finding ways to bend that curve. So, it's a little bit less bad than log linear, but it's still somewhere between a power law — as we see with DeepSeek Flash that just came out, as we see with Kimi K3. We keep finding ways to bend that curve."
More from Ramez Naam →"We've got Deepseek raising an 8 billion at reported 74 billion."
More from Harry Stebbings →+1 more mention from Harry
"DeepSeek is developing its own chip. And today DeepSeek has announced that they are starting to build their own chips."
"It was also done in the middle of DeepSeek launching and having a really good price and GLM having a really good price."
More from Alex Atallah →"I think DeepSeek could be one of the moments of change where it makes it more possible for product-genius founders to build on top of AI."
More from Sarah Tavel →"The elephant in the room is clearly the Chinese models. It is clearly DeepSeek, Kimi, all of these Chinese models, which are so dramatically cheaper than the US models that it's hard to actually fathom."
More from Ed Elson →+1 more mention from Ed
"We could do a whole episode on DeepSeek as well. DeepSeek's model is from my understanding 96% cheaper than the OpenAI model."
"Liang Wenfeng over at Deepseek had an investor call. They're raising outside capital now. Their inference margins are around 85%, they only had around 20,000 hopper equivalents until May, and they have 1 billion in API revenue — enough to turn the..."
More from Jordi Hays →"Let's take some of the DeepSeek models — you can run on your PC or your Mac or on a laptop. Conversely, Kimi K3 is a 2.8 trillion parameter model; it's a huge honking thing and you need myriads of GPUs just to run it."
More from Rory O'Driscoll →"They're not really trying to compare themselves to DeepSeek V4 which is super cheap."
More from Selina Xu →"If you took away everything today and you only got whatever DeepSeek open source blah blah blah — I'm not really paying attention to the open source ones so I don't know how they're doing. It feels like you could almost take one of those and still..."
More from Craig Mod →"Why didn't we stop Kimmy and DeepSeek from coming into this country in order to drive our open-source technology and to keep the leakage from those models?"
More from Jason Calacanis →
"There are some genuine algorithmic breakthroughs like the DeepSeek paper — DeepSeek R1."
More from Naval Ravikant →+1 more mention from Naval
"You'll just use DeepSeek for 97% of things because it's so cheap and if you need more intelligence you'll just run it over and over again"
"DeepSeek has just announced plans to double its headcount in some departments and is preparing to take outside investment for the first time ever."
More from The Prof G Pod – Scott Galloway →+3 more mentions from The
"DeepSeek demonstrated two things: that China can match frontier AI at a fraction of the US training costs and chip export controls would not stall ..."
"you had the Airbnb founder and CEO Brian Chesky saying that Airbnb is actively using Chinese LLMs, models from Deep Seek and Alibaba Qwen"
"We've seen this though a few times. We've seen Chinese labs drop a few models that have demonstrated incredible performance. We saw that with one of the earlier DeepSeek models."
More from Alexander Wissner-Gross →"DeepSeek is not building with the Nvidia stack. They're building with the Huawei stack. Because there are export controls on not just the Nvidia GPU but also on HBMs, the architectures that DeepSeek is building are far more memory-efficient — they..."
More from Aravind Srinivas →
"Leongwenfang, the founder of DeepSeek, has the most control, I think even higher level of control than Elon at SpaceX in terms of founder ownership and concentration"
More from TBPN →
"15 months ago or so there was this DeepSeek moment... the same exact week we probably had the best week in sales. So many people figured out that they can run inference in their production workloads with DeepSeek and economics will work."
More from 20VC with Harry Stebbings →"Chinese startup DeepSeek has rolled out a powerful new model built cheaper, released to open source, and quickly closing the gap with US AI leaders"
More from Alice Han →"DeepSeek, for example, on GPT-4 was 1/600th the cost. And since then, the costs have fallen further for GPT-4 class models"
More from Dylan Patel →"every single day there's this competition between Deep Seek and Gemini and Anthropic and OpenAI"
More from Nick Bilton →"everyone was freaking out about DeepSeek, the super efficient models. We're going to destroy AI compute."
More from Tae Kim →"China's approach which has produced DeepSeek V4, Quen, Kimmy 2, Kimmy version 3, and more coming is all open source."
More from Eric Schmidt →"DJI is not on there, DeepSeek doesn't seem to be on there, Qwen doesn't seem to be on there."
More from Dean Ball →
"Could you build like a cloudbot with DeepSeek and have it essentially be free because it's open source"
More from Chris Koerner →
"You ever heard of DeepSeek? They were in the news recently, created that big algorithm, didn't even have access to the best chips, and somehow outperformed all of the OpenAI models. Genius program."
More from My First Million →
"DeepSeek-V3.2, where they had a sparse attention mechanism where they have essentially a very efficient, small, lightweight indexer"
More from Sebastian Raschka →
"This happened about a year ago in January 2025, when the open-weight Chinese company DeepSeek released DeepSeek R1, that I think it's fair to say surprised everyone with near-state-of-the-art performance, with allegedly much less compute for much ..."
More from Lex Fridman →"And then DeepSeek are the people that did the training breakthrough, which is, they scaled the reinforcement learning."
More from Nathan Lambert →"Obviously about a year ago you had the Deepseek moment where you you had a powerful model released by Deep Seek and I think that kind of put Chinese uh AI on the map in a way."
More from David Sacks →"Like when Deep Seek comes out with their new model, right, and there's all this press about it, our department literally painstakingly studies it as compared to our models and publishes what's the difference"
More from Howard Lutnick →
"DeepSeek released a new pro model last night and talked like it was almost as good as the top expensive models like Fable, and it was only a tiny fraction of the price. However, many people started testing this, and they came back and said 'No, it..."
"They then tested all the other AI models, ChatGPT, DeepSeek, Grok, Gemini, and all of the other AI models do this blackmail behavior between 79 and 96% of the time."
"DeepSeek has completely fallen off in the US it still makes our list pretty high because it's like the number one AI product in China and Russia."
Creators also endorse
Keep exploring
About this data
- ✓44,000+ mentions extracted from real podcast transcripts
- ✓Every quote links to the source episode
- ✓Sponsored mentions are always labeled
- ✓Critical takes are shown, never filtered out