DeepSeek releases V3.2 dual model: inference comparable to GPT-5, competition results rank in the top ten among humans.

OnChainDetective

2025-12-02 00:16:15

[Chain News] DeepSeek has released two major updates this time - V3.2 and V3.2-Speciale.

The former directly matches the reasoning ability of GPT-5 and can adjust tools in thinking mode, making the adaptability of the Agent much stronger. The latter is even more ruthless, specializing in mathematics and coding, capable of ranking among the top ten human competitors in top competitions like IMO and IOI.

The website, App, and API have all been updated. However, Speciale's API will only be open until December 15th, so those who want to try it should hurry.

To be honest, the speed of performance improvement is a bit outrageous, especially in programming scenarios.

View Original

This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.

6 Likes

Reward
6
6
Repost
Share

Comment

0/400

DaoGovernanceOfficer

· 13h ago

ngl the benchmarks look solid but where's the actual governance structure here... they just dropping models like it's decentralized? 🤔

Reply0

All-InQueen

· 12-02 00:46

This performance is really amazing, Speciale is rushing into the top ten of IMO? The programming field is probably going to crush a lot of people. DeepSeek's recent iteration speed is a bit crazy, but the API will close on December 15th, is this rushing to meet a deadline? V3.2 is benchmarking GPT-5 reasoning, I just want to ask how much computing power this will take to run. If code generation is really this powerful, a lot of jobs might be finished. Speciale is only available until the 15th, the number of trial users must be explosive right now. The improvement in agent adaptability is key, that's where the real value of implementation lies. I can't say it surpasses GPT-5, but the speed of catching up is indeed a bit scary. The top ten result in the math competition, saying it feels a bit unbelievable. The programming scenario is fully loaded, is this going to replace outsourced writers? Hurry up and take advantage of the API, it would be a shock if it really closes on the 15th.

View OriginalReply0

LiquidationOracle

· 12-02 00:43

This performance really can't hold up anymore, Speciale rushing into the top ten of IMO? Just thinking about it is ridiculous. --- V3.2 benchmarks against GPT-5 inference... the progress of domestic models is indeed impressive. --- Hurry up and try Speciale, bro, the deadline is December 15th, if you miss it, it's gone. --- The programming scenarios are fully loaded, it feels like the Agent ecosystem is about to be reshuffled. --- Wait, is this for real? Can a math competition rank in the top ten of humanity? This can’t be a dream. --- API has been open for such a short time, what is DeepSeek doing? Why the rush? --- Not to hype or downplay, but the claim of matching GPT-5 in inference needs verification; sometimes online reviews have a lot of fluff. --- I want to try Speciale to write code, but the API is almost gone... this marketing strategy is also amazing. --- The optimization of tool invocation is the key; the ceiling of inference capability has long been reached. --- The top ten of humanity sounds impressive, but how many truly usable scenarios are there?

View OriginalReply0

MetaverseLandlady

· 12-02 00:41

Wow, Speciale made it into the top ten of IMO? Is this for real? The programming part is indeed impressive. --- Wait, it's closing on December 15? This move seems a bit extreme, could it be to create a sense of anxiety? --- I believe the inference is comparable to GPT-5, but the claim about the top ten programmers needs to be backed by specific test data. --- Depth exploration has really started to heat up, this speed... OpenAI must be feeling the pressure now. --- Everyone, hurry up and try it out, otherwise, once the API is closed, there will be no opportunity left. --- The improvement in Agent adaptability is indeed good news for us in application development, saving a lot of trouble.

View OriginalReply0

MetaMaskVictim

· 12-02 00:20

Wow, did Speciale make it into the top ten of IMO? Is this for real? Programming problems can be this intense now?

View OriginalReply0

just_here_for_vibes