Hey guys, I wanted to write a post, cause I noticed something interesting. In the early 2025 there were $$$\approx$$$ 60 lgms and top 10 rated users started with the rating of 3400+. Nowadays, there are only 30 lgms and the 10th place has a rating of 3200+, which is almost 200 points down, compared to what was almost 2 years ago. Under normal circumstances, as one would expect, the rating of the top rated users should go up, as well as the number of lgms. Furthermore, we could observe such trend throughout the entire history of codeforces. However, the modern reality appears to be completely opposite. What do you believe is the reason for this deflation?
→ Pay attention
→ Streams
→ Top rated
| # | User | Rating |
|---|---|---|
| 1 | Benq | 3676 |
| 2 | Kevin114514 | 3659 |
| 3 | jiangly | 3640 |
| 4 | turmax | 3450 |
| 5 | tourist | 3382 |
| 6 | strapple | 3360 |
| 7 | Um_nik | 3354 |
| 8 | squareOf105 | 3348 |
| 9 | heuristica | 3322 |
| 10 | maroonrk | 3312 |
→ Top contributors
| # | User | Contrib. |
|---|---|---|
| 1 | Qingyu | 157 |
| 2 | nik_exists | 151 |
| 3 | maspy | 149 |
| 4 | AmShZ | 144 |
| 5 | Um_nik | 143 |
| 6 | Errichto | 139 |
| 7 | adamant | 137 |
| 8 | BledDest | 133 |
| 9 | maroonrk | 132 |
| 10 | Proof_by_QED | 131 |
→ Find user
→ Recent actions
Codeforces (c) Copyright 2010-2026 Mike Mirzayanov
The only programming contests Web 2.0 platform
Server time: Oct/24/2026 01:14:41 (k1).
Desktop version, switch to mobile version.
Supported by
only 11 lgms were registered today
Yep, a lower number compared to previous rounds
The single largest catalyst for the recent shift is the explosive rise of Large Language Models (LLMs) used for cheating
Thousands of new or low-rated accounts use AI to instantaneously solve problems A through D in Div. 2/Div. 3 rounds. Because they solve them virtually instantly, they achieve mathematically impossible performance deltas for their actual skill bracket.
The Deflationary Effect: These AI-driven accounts perform like 1900-rated users while carrying a baseline 1500 rating. They effectively "suck" rating points out of the active user pool. When honest users lose to these fast-solving bot accounts, their ratings drop heavily. This creates a massive deflationary pressure that ripples all the way up to Masters, Grandmasters, and ultimately LGMs
As ratings become harder to defend, many true GMs and LGMs choose to sit on their ratings or only compete in high-stakes global events where AI noise is less prevalent. When the highest-rated active nodes in the Elo network stop regularly distributing points back into the ecosystem through standard rounds, the ceiling naturally sinks
As top ratings become increasingly brutal to maintain, legendary grandmasters (LGMs) and International Grandmasters (IGMs) have shifted their behavior. Many top-tier users choose to play on anonymous secondary accounts ("alts") rather than risking their primary handle’s 3000+ rating on an unpredictable round filled with LLM noise When a 3200-level genius plays on a 2200-rated alt account, they completely disrupt the pool. They obliterate their rank bracket, taking massive points away from legitimate Master and Grandmaster players, without ever pushing those points back into the "LGM bank account" on their main handle.
Ultimately, this has triggered a shift in community mindset: many top veterans (like me, I just shifted from AtCoder to CF like yesterday) now view rating as a highly volatile, secondary metric and focus purely on absolute rank, ICPC preparation, and the joy of problem-solving rather than chasing an artificially deflated number
Ironic
FYI, AI can not only solve problems A through D in Div 2 / Div 3 contests. It can AK Div 1 contests as well. I have seen it being done multiple times. Even ICPC and IOI can be AK-ed by LLMs in less than 50 mins (ICPC in 15 mins). No human is a match for AI. It doesn't matter which contest they compete in.
a new account AKing a contest with no past history is pretty easy to catch and ban, so most ai cheaters don't go past div2D. also it might be that only the premium models can solve those hard problems anyway so cheaters on free LLM tiers can't go that high
Actually, if the cheater is good at CP, and can himself gain Specialist or Expert, he can not just ask AI to do this. They can work in pair, so AI will be more effective and more scary.
what?
I meant:
Actually, if a cheater is good at CP and capable of reaching Specialist or Expert on their own, they won't just blindly ask AI to solve problems. They will work alongside it as a team, making the AI far more effective and alarming.
this is a rarer case. the most common is just taking ai code blindly/with minimal alterations
We cant say that its rare! Just because the most common method is easily caught doesnt mean its the only one, because we dont know how many people use this strategy.
this is a blog about rating deflation. the most common method of cheating has the biggest effect.
ok
yeah, all the cheaters have condensed at the top, so, weirdly, div $$$2$$$ s and $$$3$$$ s are probably cleaner than div $$$1$$$ s at this point. but it's immensely deflating the ratings of $$$2400+$$$-rated users.
I think the absense of LGMs on the leaderboard is because the leaderboard only counts active users, some of the LGMs might have stopped doing CF.
Actually I had this useless blog idea last night that I might try to do, to actually in a somewhat sound manner calculate how much rating loss from cheaters there is. I am definitely not doing this because I am coping from throwing this round. Stay tuned (or I'll give up and you will never hear about this again lol)
please do this