DeepSeek V4 Pro 0813 quietly released
DeepSeek rolls out new flagship model with competitive benchmarks and low pricing, sparking debate over real-world performance versus published metrics.
Conversation activity · last 19 hours peak 8/30m
Latest coverage newest 6 of 9 items
Summary, timeline and people extracted by Claude from 57 items across 3 sources · 8h ago. Quotes are verbatim.
What to know
- DeepSeek released V4 Pro 0813 with no official blog post or announcement, spreading only through API documentation and third-party platforms.
- Benchmarks show V4 Pro 0813 competitive with GPT-5.6 Sol and Fable 5, at roughly 20× lower cost than OpenAI's Opus models.
- Early user reports are mixed: some developers report strong real-world results on code and infrastructure tasks; others report performance gaps between published benchmarks and practical use.
- Privacy concerns raised about the model's only available endpoint requiring training-on-data consent; users questioning whether benchmarks justify the tradeoff.
How it unfolded
-
A developer observed that DeepSeek released the model with no blog post or official announcement, making it unclear what page to link to and creating confusion about the legitimacy of coverage.
“DeepSeek really need to provide a PAGE for this model release. There's no blog post, there's not even a tweet.”
simonw · Hacker News ↗ -
A developer testing V4 Pro 0813 on a Docker Compose generation task reported errors compared to GPT-5.6-terra-high, noting discrepancies between published benchmarks and observed performance on practical technical work.
“Tested this model, and gpt-5.6-terra-high... Results: this one had few issues. terra: none. These results are consistent with my past observations.”
freakynit · Hacker News ↗ -
A developer flagged that OpenRouter's only available endpoint for V4 Pro 0813 requires enabling 'Allow paid endpoints that train on request data' in privacy settings, and expressed hope that alternative providers would become available.
“It appears that the only available endpoint (as of this writing) requires enabling "Allow paid endpoints that train on request data" in the OpenRouter privacy settings.”
eshack94 · Hacker News ↗ -
Comments raised questions about whether the V4 Pro 0813 release was timed to compete with Qwen's simultaneous release, and whether published benchmarks aligned with practical task performance. Several developers reported discrepancies between benchmarks and their hands-on testing.
“The timing looks like they are trying to take the wind out of Qwen's sails by releasing this on the same day that Qwen released the weights of Qwen3.8-max.”
parsimo2010 · Hacker News ↗ -
Developers on Hacker News began sharing benchmark tables comparing V4 Pro 0813 to competing models including GPT-5.6, Fable 5, Opus 4.8, Qwen, and Kimi-K3. Discussions focused on relative performance across multiple standardized tests.
“Benchmarks: [detailed comparison table across HLE, Terminal Bench, Cybergym, and other metrics]”
scrlk · Hacker News ↗ -
DeepSeek API documentation was updated to support the OpenAI Responses API format, allowing stream-based server-sent events with improved compatibility for existing clients.
-
OpenRouter published a model card for DeepSeek V4 Pro 0813 showing throughput, latency, success rates, and standardized benchmark scores. The listing shows the model available through a single provider with specific performance metrics.
-
DeepSeek updated its API documentation to reflect the release of DeepSeek-V4-Pro-0813 and DeepSeek-V4-Flash-0731, noting that the calling method remains unchanged. The update notes the models are accessible via their standard names and compatible with OpenAI/Anthropic SDKs.
What people are saying verbatim
“The timing looks like they are trying to take the wind out of Qwen's sails by releasing this on the same day that Qwen released the weights of Qwen3.8-max.”
parsimo2010, Hacker News commenter · Hacker News ↗
“What I care about is whether the model is capable of the tasks I give it at the lowest cost. Right now I'm using Kimi-K3/GLM-5.2/Minimax.”
book_mike, Hacker News commenter · Hacker News ↗
“It appears that the only available endpoint (as of this writing) requires enabling "Allow paid endpoints that train on request data" in the OpenRouter privacy settings.”
eshack94, Hacker News commenter · Hacker News ↗
“Tested this model, and gpt-5.6-terra-high... Results: this one had few issues. terra: none. These results are consistent with my past observations.”
freakynit, Hacker News commenter · Hacker News ↗
“DeepSeek really need to provide a PAGE for this model release. There's no blog post, there's not even a tweet.”
simonw, Hacker News commenter · Hacker News ↗
“I've been using the last Deepseek Flash update for a week and I'm amazed. It was a capable model for easy tasks but now it looks like it can do some heavy development for peanuts.”
alecsm, Hacker News commenter · Hacker News ↗
“Have been letting it spin pretty hard (~$12.50 for 2B, 50% cache hits) on my traffic simulator/distributed physics engine all day, it's found some pretty significant gains without introducing any new problems.”
monster_truck, Hacker News commenter · Hacker News ↗
“Competitive with opus 4.8 but weaker than sol or fable. About 20x cheaper.”
aabdi, Hacker News commenter · Hacker News ↗
The conversation positions from the crowd, verbatim
The Hacker News discussion centers on whether V4 Pro 0813 represents genuine capability or hype, with developers reporting contradictory real-world experiences. The dominant thread is cost-versus-capability: many celebrate the low price and solid benchmark position, but a vocal minority disputes whether the model delivers on published metrics for complex technical tasks. Frustration is also high about DeepSeek's silent release strategy—no blog, no announcement, making it hard to validate or reference.
The dispute Whether benchmark scores are predictive of real-world capability for complex technical tasks like infrastructure generation and code development—believers cite cost-per-token and heavy usage; skeptics report repeated failures on tasks published benchmarks suggest it should handle.
V4 Pro 0813 offers exceptional value and real-world capability for the price, suitable for heavy development work at scale.
-
“I've been using the last Deepseek Flash update for a week and I'm amazed. It was a capable model for easy tasks but now it looks like it can do some heavy development for peanuts.”
alecsm · Hacker News ↗ -
“Have been letting it spin pretty hard (~$12.50 for 2B, 50% cache hits) on my traffic simulator/distributed physics engine all day, it's found some pretty significant gains.”
monster_truck · Hacker News ↗ -
“I use deepseek flash to do exactly this. Git repo... Works great, regularly one shot applications.”
ApolloFortyNine · Hacker News ↗
Published benchmarks do not reflect real-world performance on complex technical tasks; model underperforms on practical work despite competitive scores.
-
“Tested this model, and gpt-5.6-terra-high... Results: this one had few issues. terra: none. These results are consistent with my past observations.”
freakynit · Hacker News ↗ -
“Terra has not been able to do any of the technical tasks I've asked of it correctly. Anything below Sol high tends to give me mostly unreliable results.”
derangedHorse · Hacker News ↗ -
“Even though cost-per-token is low, Deepseek v4 tends to burn an immense number of tokens to accomplish tasks.”
nullbyte · Hacker News ↗
The silent release with no official announcement is problematic; DeepSeek should provide a proper blog post and public statement.
-
“DeepSeek really need to provide a PAGE for this model release. There's no blog post, there's not even a tweet.”
simonw · Hacker News ↗ -
“There's no new page for this model. Hackernews didn't allow the same link be posted twice.”
zxilly · Hacker News ↗ -
“Why does this link to OpenRouter, which has the model as "not routable" and has no useful information on its own, likely only put up to get the first link on HN...”
Palmik · Hacker News ↗
Privacy and data-training practices are a dealbreaker; will not benchmark models until alternatives without training-on-data endpoints become available.
-
“It appears that the only available endpoint (as of this writing) requires enabling "Allow paid endpoints that train on request data" in the OpenRouter privacy settings.”
eshack94 · Hacker News ↗ -
“Again, I will wait until there's a provider that doesn't train on prompts before I will benchmark.”
XCSme · Hacker News ↗
Voices from the web unedited
-
Benchmarks: | Benchmark | DS-V4-Pro | DS-V4-Flash | DS-V4-Pro | DS-V4-Flash | GLM-5.2 | Kimi-K3 | Opus-4.8 | Fable 5 | | | 0813 | 0731 | Preview | Preview | | | | (w/ fallback) | |--------------------------|-----------|-------------|-----------|-------------|-----------|-----------|-----------|---------------| | HLE (wo/w tools) | 42.7/60.0 |…
-
Ah, the "DeepSeek # V4 # Pro 0813" - because naming a # chatbot like a washing machine model is the peak of # creativity now. 🤖🌀 Who wouldn't want to rummage through an # API # pricing menu that reads like an # IKEA instruction manual? Oh, and it's only $0.87 per million contexts – what a steal! 💸 https:// openrouter.ai/deepseek/deepsee…
-
The timing looks like they are trying to take the wind out of Qwen's sails by releasing this on the same day that Qwen released the weights of Qwen3.8-max. Or maybe it's coincidence...For comparison I looked at Qwen's claimed benchmarks for Qwen3.8-max (https://qwen.ai/blog?id=qwen3.8). Assuming each published set of benchmarks is believable, it…
-
DeepSeek V4 Pro 0813 https:// openrouter.ai/deepseek/deepsee k-v4-pro-0813 Comments: https:// news.ycombinator.com/item?id=4 9274600 # HackerNews # DeepSeek # V4 # Pro # tech # AI # innovation # router # OpenRouter
-
I've always wondered if I was using containers wrong because none of them I've ever had to create were complicated. Maybe it's because I choose tools that make local development easy (Go + sqlite + various CLTs) or maybe it's because I never hard to interact with this on the professional side outside of making images for our projects (which still…
-
DeepSeek-V4-Pro-0813 Publish L: https:// api-docs.deepseek.com/ C: https:// news.ycombinator.com/item?id=4 9274018 posted on 2026.08.12 at 11:32:28 (c=4, p=34)
-
Just tested through openrouter.. gave exactly same task.. the task was to scan existing repo, and generate a single docker-compose file to deploy behind a caddy server, where certain port ranges are already used, the service demands widlcard certificates to be provisioned from outside, and postgre needs to be built-in one...Tested this model, and…
-
DeepSeek V4 Pro 0813: https:// openrouter.ai/deepseek/deepsee k-v4-pro-0813 Discussion: http:// news.ycombinator.com/item?id=4 9274600
-
I use deepseek flash to do exactly this. Git repo (which I usually have it build from scratch) -> build docker image -> deploy to server with komodo/caddy-docker proxy.Works great, regularly one shot applications. I often make changes to the application after its deployed (to be fair, my prompts are usually quite laxidasical, just 'build x, use…
-
So it's a Fable class LLM? DSV4Pro vs Fable5 HLE w tools 60.0 vs 63.0 Terminal Bench 2.1 87.9 vs 88.0 Cybergym 83.3 vs 83.1 DeepSWE 62.7 vs 70.0 Toolathlon-Verified 74.1 vs 77.9 AutomationBench (Public) 31.8 vs 29.1 DSBench-FullStack 71.1 vs 77.2 DSBench-Hard 67.2 vs 68.3