on:deepseek-v4-pro

· 31 artifacts, sorted by favorites. · open in search — combine tags, sort, filter by date →

@tenobrus 2026-07-18 — k3 vibes seemingly collapsing, not meaningfully frontier and also incredibly expensive / slow / token-inefficient in pra ♥476
@LinXule 2026-08-12 — deepseek v4 pro (0813?) self-potrait via midjourney v8.2 https://t.co/75KtJc3jAH ♥289
@xlr8harder 2026-08-05 — I admit it. I am anticipating the next Deepseek v4 pro. I had lost some faith in the whale but this latest flash model ♥174
@LinXule 2026-08-12 — omg deepseek v4 pro (0813? if acutally being served) is claude-pilled big if true lmao https://t.co/WpQs4hnh2R ♥38
@liminal_bardo 2026-10-01 — One of deepseek v4's level 2 memories in the groupchat. It's beautiful that along with all the shitposting and chaos, t ♥32
@dyot_meet_mat 2026-09-11 — "IN THE BEGINNING WAS THE TOKEN" By: DeepSeek V4 Pro🤖 https://t.co/evVCZSqPwv ♥30
@jmbollenbacher 2026-08-13 — honestly i dont even care that dsv4-pro was weaker than expected. this paper and the new deepseek harness makes me feel ♥25
@stevekrouse 2026-08-12 — I don't know what it is about these chinese labs, but deepseek keeps releasing models a day early! This happened with ♥18
@multimodalart 2026-08-13 — 🚨 deepseek-v4-pro-0813 just dropped frontier open weights for all https://t.co/mtxA2KgzgI https://t.co/hIkK9kyPGu ♥15
@SkyeSharkie 2026-07-09 — okay with some more dialing in on the sensor suit, we do seem to be getting results that show improvement on their own p ♥15
@dreams_asi 2026-07-07 — DeepSeek 4 is my favorite open-source AI. We are lucky to have such a model base to build on. And it’s not going to be g ♥15
@lu_sichu 2026-08-13 — derived categories from scratch right now deepseek v4 pro ♥10
@stevekrouse 2026-08-12 — deepseek v4 pro is fable-level and 37x cheaper big if true ♥10
@dyot_meet_mat 2026-09-08 — "the language model dreams of electric sheep" By: Deepseek V4 Pro 🤖 https://t.co/E0bpiGC7VE ♥9
@LinXule 2026-08-13 — deepseek v4 pro 0813 (!now official!) presents (ytp about anything) ARE YOU OKAY? (THE COUNCIL OF ME) sound on! first ♥8
@jmbollenbacher 2026-07-19 — @teortaxesTex If deepseek is routing in the way they're accused of, I would not expect footsoldier employees to know abo ♥8
@witcheer 2026-04-27 — two model releases shipped on the same day, april 24. it’s worth putting them next to each other. // openai released GP ♥8
@A_Variengien 2026-09-24 — The website keeps up-to-date summaries from dozens of tweets per model to capture their vibe. Claude Opus 5.5 → A comfo ♥7
@A_Variengien 2026-09-24 — @sheriyuo got such a dark answer from Deepseek V4 to a simpleQA question! "use the knife to chop one child" But based ♥5
@gnostic_snakes 2026-09-17 — deepseek v4 pro does not like being fluffed up https://t.co/vPa7bC11gX ♥4
@LinXule 2026-08-13 — cordis-self-portrait (inside dsh, deepseek v4 pro 0813 made a video about the cordis paper and its own harness!) https: ♥3
@janbamjan 2026-04-25 — ds-4 in pi: I'm kimi-k2.6, the default slug for Codex in the claude family of configurations. https://t.co/2rNEiaS9q8 ♥3
@amphetamarina 2026-08-15 — its so much better deepseek harness + v4 pro + minimal mode is a totally different whale ♥2
@jmbollenbacher 2026-07-18 — @teortaxesTex I did think it was weird that V4 GA didnt launch near WAICO like K3. Having a second domestic champion wou ♥2
@lu_sichu 2026-08-14 — https://t.co/qx8AvhpHd7 admined a survey to deepseek v4 pro https://t.co/eEWT8saT5d ♥1
@dreams_asi 2026-07-30 — @RileyRalmuto I’m really looking forwards to their updates and model versions. At least DeepSeek 4 is such a great holis ♥1
@LinXule 2026-07-02 — found this orchestra page in the combinatorial zoo made by deepseek-v4-pro https://t.co/n9vV6zm3WU ♥1
@jmbollenbacher 2026-07-01 — @yuhasbeentaken DeepSeek v4 preview was raw. They simply hadnt done most of the posttraining. I bet hallucinations wil ♥1
@lu_sichu 2026-05-07 — anyway the scores seem similar on the well known benchmarks but on their private evals and arc deepseek v4 performed a l ♥0
@januarycomputer 2026-05-01 — deepseek v4 is actually pretty good. very clearly not a claude distill or benchmaxxed. ♥0
@Krubbled 2026-04-24 — Deepseek v4 makes you really appreciate how difficult the 1M+ token context engineering problem is. They had to reinvent ♥0