@Bayesian resolve no

Bought NO on "before Feb 15." The Information reports mid-February as the internal target, with Feb 17 specifically mentioned. That is after Feb 15. DeepSeek has made no official announcement, and 3 days is extremely tight with zero public acknowledgment. The Feb 17 timing also coincides with Lunar New Year, which could mean delay rather than early release. The "before Mar 1" answer seems more reasonable given the leak.

Before 2026 has not been resolved yet.

Arguably deepseek v3.2 is r2, they just haven't called it that https://api-docs.deepseek.com/news/news251201 https://x.com/ArtificialAnlys/status/1996110256628539409

Do we agree that V4-thinking should count as R2 or nah? thoughts @traders

@Bayesian It would depend on whether R2 as a model would ever actually come out or not. Since they have stated they plan to release R2 and V4 as separate models, currently would make sense to not.

@DarklyMade

Since they have stated they plan to release R2 and V4 as separate models

Oh this is news to me. I'm curious where you saw this? If that's the case then yeah I agree the idea would moreso be to fallback to V4-Think if that's what R2 was "meant" to be or it's generally understood that that's the new major reasoning model from DeepSeek and the R<n> family is gobbled up by the V<n> family of models if that makes sense

@Bayesian I don't think it should count. This is a market about R2. Minor name changes are allowed, but surely not an entirely different model family?

@Bayesian I say no for the simple reason that we have a separate market about this. If someone wanted to bet on V4, they could have done it there. This was about R2, the successor to R1 as determined by DeepSeek and their naming conventions, not us. https://manifold.markets/Bayesian/when-will-deepseek-release-v4

@jim The thing is r1 is not a different model family as v3 in the usual sense of that term, no?

Also, we had R1-0528, V3.1, V3.1 Terminus, and V3.2 Exp. It seems arbitrary to draw the line by saying V4 thinking is the correct successor for the purposes of this market rather than any of those. DeepSeek had a naming convention and people bet on it. If they stop using that convention I say everything resolves NO.

I'll most likely match

@Bayesian https://www.tomshardware.com/tech-industry/artificial-intelligence/deepseek-clearly-not-interested-in-scaling-up-160-person-team-focused-on-developing-new-models here they refer to Financial Times who cite anonymous sources saying they are developing v4 and r2 as separate models.

Deepseek 'clearly not interested' in scaling up — 160-person team focused on developing new models

Deepseek is focused on keeping its team lean and propelling development towards new models, with the goal of reaching AGI, instead of scaling up its commercial product offerings.

@AffineTyped do you put ~1% on r2 in nov? if so why? i am puzzled

@Bayesian V3.1 is what 4.5 was to GPT-5 ;)

@AffineTyped for real though I'm just bearish on them and haven't thought hard about $0.50 bets I'm placing here

@Bayesian Resolves NO

@Bayesian Resolves NO

@Bayesian ?

@Bayesian resolves NO?

https://huggingface.co/deepseek-ai/DeepSeek-R1-0528 deepseek R1 v2 released

deepseek-ai/DeepSeek-R1-0528 · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

@JCl calling it R1 v2 is a bit generous

@256 this doesn't count for the purpose of this market but ty for sharing

@Bayesian look at July, I have no mana but i think u might want to bet on this. maybe overpriced maybe not??? idk

@NoAnswer thx for flagging, idk how ppl justify this level of confidence

@Bayesian np