dhuamanilu's blog

Gemini 3 Deep Think claims 3455 Elo on Codeforces (Legendary Grandmaster?)
By dhuamanilu, history, 8 months ago, In English

Just saw the news regarding the release of Gemini 3 Deep Think (12 Feb 2026) and the benchmarks are getting crazy.

According to their latest technical report, this new model achieves a Codeforces Elo of 3455.

That puts it deep into Legendary Grandmaster territory. What makes this more impressive is that the benchmark specifies this was achieved with "No tools", meaning it's raw generation without code execution support during the contest simulation.

Has anyone with early access tested it on recent Div 1 rounds?

  • Vote: I like it
  • +70
  • Vote: I do not like it

»
8 months ago, hide # |
 
Vote: I like it +7 Vote: I do not like it

Auto comment: topic has been updated by dhuamanilu (previous revision, new revision, compare).

→ Reply
»
8 months ago, hide # |
 
Vote: I like it -21 Vote: I do not like it

In fact, I found that someone in Linux Do community has already built an agent with Gemini 3 flash, which can AK global round 31.

→ Reply
»
8 months ago, hide # |
 
Vote: I like it +24 Vote: I do not like it

we r cooked

→ Reply
»
8 months ago, hide # |
← Rev. 2  
Vote: I like it 0 Vote: I do not like it

We've seen people reaching 2500, cheating with LLMs (Gemini 3 Pro?), so it sounds legit.

→ Reply
»
8 months ago, hide # |
← Rev. 2  
Vote: I like it 0 Vote: I do not like it

What to hell is that? First, thats cool that people can do such a powerfull AI, secondly, thats bad, because:

if (AI better) more cheaters;
→ Reply
  • »
    »
    8 months ago, hide # ^ |
     
    Vote: I like it +6 Vote: I do not like it

    This is not precise. It should be:

    if (_free or cheap_ AI better) more cheaters;

    I don't think a deep think model is free or cheap.

    → Reply
    • »
      »
      »
      8 months ago, hide # ^ |
       
      Vote: I like it +19 Vote: I do not like it 
      if (_free or cheap_ AI better) more cheaters;
else if (gemini_ultra_cheaters.size() >= 1) {
    gemini_ultra_cheaters[0].leak_solutions();
    more cheaters;
}
      → Reply
    • »
      »
      »
      8 months ago, hide # ^ |
      ← Rev. 3  
      Vote: I like it +1 Vote: I do not like it 
      if (expensive AI better) sleep(6 months);
assert(free or cheap AI better);
      → Reply
»
8 months ago, hide # |
 
Vote: I like it 0 Vote: I do not like it

i think such AI at least can be helpful with understanding some hard topics .But i am curious about some gemini 10 for example . It can be actually the best int the world , and then it will be totally about who can use AI better .

→ Reply