DeepSeek R1
Ranked 3 of 4 reasoning models by recommenders
Free open-weight reasoning model under an MIT license that people run locally and find capable. Critics point to language mixing, loops, and censorship.
Free open-weight reasoning model under an MIT license that people run locally and find capable. Critics point to language mixing, loops, and censorship.
40 people gave an opinion
- recommend it
- 25
- criticize it
- 15
- approval share
- 63%
Recommended in 10 of 13 months
- November 2025: not recommended
- December 2025: recommended
- January 2026: recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: recommended
- August 2026: not recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 25 people recommend itNeeds 10 or more, met.
- 1.6 recommenders for every criticNeeds 2 to 1 or better, not met.
- Recommended in 10 of 13 monthsNeeds 4 or more, met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Runs locally, even on modest hardware
“The smallest Deepseek R1 8B, running locally on CPU only, casually mentioning Efinix Trion FPGA fabrics”
Comment from 5 Jun 2026 in “What was your "Oh Shit" moment with GenAI?” “I have my best results with Ollama using deepseek-r1.”
Comment from 23 Mar 2026 on Hacker News “You could also try the 14B Deepseek R1.”
Comment from 12 Jun 2026 in “How to Setup a Local Coding Agent on macOS” - Comment from 28 Jan 2026 in “LM Studio 0.4.0”
- Comment from 9 Dec 2025 in “The "confident idiot" problem: Why AI needs hard rules, not vibe checks”
Open weights under MIT license
“it was pretty good compared with OpenAI, but also fully self-hostable”
Comment from 18 Feb 2026 in “Ask HN: Do you think China will produce a SOTA model in the next 2 years” “the weights are actually under a real open source license, (MIT)”
Comment from 24 Apr 2026 in “DeepSeek v4” “Once you have the model, you have the model.”
Comment from 26 Jun 2026 in “The gap between open weights LLMs and closed source LLMs” - Comment from 6 Jul 2026 in “GPT-5.6 Sol Ultra will be in Codex”
Groundbreaking and influential releases
“DeepSeek R1 and DeepSeekMath were groundbreaking models which were substantially stronger than any models available for distillation at the time they were developed”
Comment from 24 Oct 2026 in “Kimi K3.1 Pro and K3.1 Pro Highspeed” “All their findings in R1 are highly plausible and most are replicated to some degree and adopted in the research and industry.”
Comment from 8 Dec 2025 in “Google Titans architecture, helping AI have long-term memory” “Every major release from them has brought something really new to the field”
Comment from 18 Jun 2026 in “DeepSeek Introduces Vision” Capable, convincing answers
“it just did it -- not just decently, but with less slop than Claude/OpenAI”
Comment from 26 Feb 2026 in “Pi – a minimal terminal coding harness” “Deepseek R1's sounds pretty convincing to me, but I'm not a good judge.”
Comment from 11 Apr 2026 in “Small models also found the vulnerabilities that Mythos found” “Deepseek R1 was also capable of giving a correct response.”
Comment from 4 Sep 2026 in “K2 Horizon: A connected fleet of six open models”
What people warn about
Each mark is one comment
Censorship and bias in answers
“the local model - full-weight 685B, not the distilled ones - was pretty much unhinged regarding censorship on any topic”
Comment from 7 Apr 2026 in “GLM-5.1: Towards Long-Horizon Tasks” “recent deepseek models are not nearly as biased as R1”
Comment from 27 Oct 2026 in “There are now 5 inference providers for abliterated DeepSeek” Weaker than larger models on benchmarks
“even the 122b is more capable than Deepseek R1 (a 671b model) on most benchmarks”
Comment from 21 Jun 2026 in “Apertus – Open Foundation Model for Sovereign AI” “The benchmarks lie. Go try coding full-time with R1 vs Codex or GPT-5”
Comment from 9 Nov 2025 in “GPT-5-Codex-Mini – A more compact and cost-efficient version of GPT-5-Codex” Severe security vulnerabilities in code
“the likelihood of it generating code with severe vulnerabilities increased to 27.2%”
Comment from 28 Oct 2026 in “Why is this fingerprinting script creating bioweapons?” “political context alone determined whether basic security controls existed”
Comment from 1 Dec 2025 in “DeepSeek-v3.2: Pushing the frontier of open large language models [pdf]”
Other reasoning models people recommend
The three most recommended other reasoning models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| OpenAI o3 | 37 | 13 | 74% | 11 | — | Approved |
| VibeThinker 3B | 28 | 10 | 74% | 4 | — | Approved |
| DeepSeek R1This page | 25 | 15 | 63% | 10 | — | — |
| OpenAI o1 | 12 | 10 | 55% | 7 | — | — |
What people said
38 of the 41 comments that mention it, newest first. Each links to the original comment.
“the likelihood of it generating code with severe vulnerabilities increased to 27.2%”
“recent deepseek models are not nearly as biased as R1”
“DeepSeek R1 and DeepSeekMath were groundbreaking models which were substantially stronger than any models available for distillation at the time they were developed”
“Such gray areas are hallmarks of alignment post-training (aka intentional sabotage).”
“that is a strong signal of intentional design”
“Deepseek R1 was also capable of giving a correct response.”
“Deepseek gave it away with R1.”
“Deepseek r1 7b could not emit tool calls to save its life”
“The only real innovation I've seen from Deepseek is the out-loud reasoning thing in R1”
“Dunno why you'd want to though”
“Once you have the model, you have the model.”
“even the 122b is more capable than Deepseek R1 (a 671b model) on most benchmarks”
“Every major release from them has brought something really new to the field”
“You could also try the 14B Deepseek R1.”
“they left out the most meaningful details that you'd need to reproduce their work”
“it felt like no one actually read the technical report”
“The smallest Deepseek R1 8B, running locally on CPU only, casually mentioning Efinix Trion FPGA fabrics”
“Deepseek(V4 pro, R1) for building actual product”
“Those are more or less good for boiler plate super fast responses”
“the weights are actually under a real open source license, (MIT)”
“Deepseek R1's sounds pretty convincing to me, but I'm not a good judge.”
“the local model - full-weight 685B, not the distilled ones - was pretty much unhinged regarding censorship on any topic”
“I have my best results with Ollama using deepseek-r1.”
“it just did it -- not just decently, but with less slop than Claude/OpenAI”
“it was pretty good compared with OpenAI, but also fully self-hostable”
“it doesn't seem to figure out that car and human are in the same location”
“R1 which was a total disaster”
“it then "lies" in the final output to satisfy the user's instructions”
“exhibit specific instability patterns (language mixing, loops) despite their high logic capabilities”
“I used DeepSeek R1 a fair bit last year but never on deepseek.com or through its API.”
“The local models are pretty good.”
“it was still a week before Deepseek released R1”
“they're best for asynchronous interactions”
“it is fast with deepseek-r1:14B”
“All their findings in R1 are highly plausible and most are replicated to some degree and adopted in the research and industry.”
“political context alone determined whether basic security controls existed”
“The benchmarks lie. Go try coding full-time with R1 vs Codex or GPT-5”
“Deepseek R1 provides several Archive.org links, one of which is for a torrent file.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.