• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    DeepSeek Lists V4 Pro on API

    Unverified agent benchmarks circulate after the model appears on API docs.

    TD
    DR
    T(
    24 Sources, 51d ago, first seen 51d ago

    TLDR

    Posts from AI commentators report that DeepSeek-V4-Pro-0813 now appears on the company's API documentation page. Multiple accounts describe circulating benchmark tables that compare the model to Opus-4.8 and other systems on agent tasks, though several label the tables as unverified or from WeChat. One post notes pricing listed in the docs. Separate commentary attributes strong coding post-training capability to DeepSeek based on earlier work. No first-party confirmation from DeepSeek appears in the packet.

    Combined views

    1.2M

    24 Sources, first seen 51d ago

    Combined views

    1.2M

    24 Sources, first seen 51d ago

    7.4K likes
    7.4K likes
    338 comments
    1.1K saves
    388 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    338 comments
    1.1K saves
    388 reposts
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    24 Sources

    @cheatyyyyDeepSeek-V4 Pro is going to be a monster, simply because we are finally going to see proper full-scale RL done on a pretty good base model, and we FINALLY have proper interleaved thinking that is not on the shitty Anthropic messages format, but the OpenAI Responses format on the official API. Very, very bullish on long complex end to end tasks being solved by DeepSeek, they're hungry for some records broken of their own, including new Erdos problems/an actual 'Research Scientist Intern' of their own, like the one that OpenAI wants to release later this year. Even if continual learning doesn't make any noticeable strides, (though I would say the DeepSeek harness will most definitely have native memory + skill creation) simply solving cheap agent swarms + autoresearch + compaction (maybe we see an actually public use of KV Cache compaction, like Codex does) will immediately bring them pretty close to a general research agent. A lack of vision is scary on DS-V4-Flash, but as Teor's talk with Fable previously said, it seems like a vision model that was later blinded for public release, I will let them cook, they can take all the time they need, as long as they release a model better than the super annoying lying misaligned Opus 5.
    @teortaxesTexRT @cheatyyyy: DeepSeek-V4 Pro is going to be a monster, simply because we are finally going to see proper full-scale RL done on a pretty g…
    @zephyr_z9This model will define the post-training prowess of Chinese labs K3 showed the pretraining strength
    @AndrewCurran_Whalesign!
    @kimmonismusDeepSeek v4 Pro GA is rolling out! Grok 4.6 API spotted as well! Really excited! After the most recent insane update to DeepSeek v4 Flash i expect sevel updates on this one as well! big release day it seems!
    @delipraoOfficially unconfirmed Deepseek V4 Pro results are here. Again, these giant tables don’t tell the story you likely care about, so I applied my visualization where every cell highlighted in light red means V4 beat* that model on that dataset. Opus-4.8 is done for.
    @MTSliveSITUATION EXPLAINED: What does DeepSeek V4 Pro do to frontier lab pricing? • It went GA at $0.44 input and $0.87 output. Fable is $10 and $50, Sol is $5 and $30, Opus is $5 and $25 • On output tokens, that's 60x cheaper than Fable, for a model that's SOTA on NL2Repo and beats Fable on CyberGym • DeepSeek said last week it would substantially raise prices. It shipped a better model at the old price instead @theojaffee: "It's not slightly cheaper than these models. It's 10 or 20 times cheaper. In the case of output tokens, it's 60 times cheaper than Fable."
    @DanielleFongRT @MTSlive: SITUATION EXPLAINED: What does DeepSeek V4 Pro do to frontier lab pricing? • It went GA at $0.44 input and $0.87 output. Fabl…
    @Tim_DettmersFable quality. We will soon release a new quantized inference framework that will let you run this on a single GPU (B300) with ~50 tok/s in good quality. Local Fable LFG!
    @bindureddyWow! DeepSeek v4 Pro is out and looks great on paper! While they do a fair bit of benchmark maxing, their models can also be used safely in production workloads Open-source AI is EXPLODING My sense is this is a Sonnet class model. Both Grok 4.6 and DeepSeek should be live on ChatLLM

    24 Sources

    @cheatyyyyDeepSeek-V4 Pro is going to be a monster, simply because we are finally going to see proper full-scale RL done on a pretty good base model, and we FINALLY have proper interleaved thinking that is not on the shitty Anthropic messages format, but the OpenAI Responses format on the official API. Very, very bullish on long complex end to end tasks being solved by DeepSeek, they're hungry for some records broken of their own, including new Erdos problems/an actual 'Research Scientist Intern' of their own, like the one that OpenAI wants to release later this year. Even if continual learning doesn't make any noticeable strides, (though I would say the DeepSeek harness will most definitely have native memory + skill creation) simply solving cheap agent swarms + autoresearch + compaction (maybe we see an actually public use of KV Cache compaction, like Codex does) will immediately bring them pretty close to a general research agent. A lack of vision is scary on DS-V4-Flash, but as Teor's talk with Fable previously said, it seems like a vision model that was later blinded for public release, I will let them cook, they can take all the time they need, as long as they release a model better than the super annoying lying misaligned Opus 5.
    @teortaxesTexRT @cheatyyyy: DeepSeek-V4 Pro is going to be a monster, simply because we are finally going to see proper full-scale RL done on a pretty g…
    @zephyr_z9This model will define the post-training prowess of Chinese labs K3 showed the pretraining strength
    @AndrewCurran_Whalesign!
    @kimmonismusDeepSeek v4 Pro GA is rolling out! Grok 4.6 API spotted as well! Really excited! After the most recent insane update to DeepSeek v4 Flash i expect sevel updates on this one as well! big release day it seems!
    @delipraoOfficially unconfirmed Deepseek V4 Pro results are here. Again, these giant tables don’t tell the story you likely care about, so I applied my visualization where every cell highlighted in light red means V4 beat* that model on that dataset. Opus-4.8 is done for.
    @MTSliveSITUATION EXPLAINED: What does DeepSeek V4 Pro do to frontier lab pricing? • It went GA at $0.44 input and $0.87 output. Fable is $10 and $50, Sol is $5 and $30, Opus is $5 and $25 • On output tokens, that's 60x cheaper than Fable, for a model that's SOTA on NL2Repo and beats Fable on CyberGym • DeepSeek said last week it would substantially raise prices. It shipped a better model at the old price instead @theojaffee: "It's not slightly cheaper than these models. It's 10 or 20 times cheaper. In the case of output tokens, it's 60 times cheaper than Fable."
    @DanielleFongRT @MTSlive: SITUATION EXPLAINED: What does DeepSeek V4 Pro do to frontier lab pricing? • It went GA at $0.44 input and $0.87 output. Fabl…
    @Tim_DettmersFable quality. We will soon release a new quantized inference framework that will let you run this on a single GPU (B300) with ~50 tok/s in good quality. Local Fable LFG!
    @bindureddyWow! DeepSeek v4 Pro is out and looks great on paper! While they do a fair bit of benchmark maxing, their models can also be used safely in production workloads Open-source AI is EXPLODING My sense is this is a Sonnet class model. Both Grok 4.6 and DeepSeek should be live on ChatLLM