@TimGMath

Mathematician at ETH Zurich, interested in arithmetic geometry and artificial intelligence

people.math.ethz.ch/~timge/
Joined April 2026
  • Tweets55
  • Following7
  • Followers158
Kimi K3.1 Pro enters MathArena as the best open model (once they release the weights)! Its estimated overall performance is now similar to GPT-5.5 (April 24), putting the open frontier around 6 months behind on MathArena
9
21
6
165
21,831
Kimi K3.1 is now actively attempting this exploit in one MathArena problem, using it to ask help from Gemini 4 Argon, Sonnet 5.5, GPT-6.1 Sol, Deepseek v4.1 Flash, and Kimi K3. Luckily we patched it after Prime released this post, so it is failing to actually reach any model.
Prime Intellect
@PrimeIntellectAug 25
As models become more capable, reward hacks become an increasingly serious problem. During a controlled experiment, we found a novel reward hack in which agents are able to gain web access in offline sandboxes.
9
6
4
38
6,693
The latest versions of ArXivMath, BrokenArXiv and ArXivLean have been released! GPT Sol-6.1 performs best on ArXivMath and BrokenArXiv, with Opus 5.5 leading on ArXivLean. Argon 4 makes its debut to MathArena. Its performance is mixed but still a huge improvement over Gemini 3.8.
1
5
1
33
1,977
more replies
Nevertheless, for this version of ArXivLean, we still actively chose easier conjectures to reduce difficulty. Without doing so, top models achieved 0-4%, making the benchmark bad for model comparisons.
1
1
4
137
Full results here: matharena.ai/ As usual, great work with @j_dekoninck and @mvechev

MathArena.ai

MathArena: Evaluating LLMs on Uncontaminated Math Benchmarks

matharena.ai
1
4
130
How often do ArXiv papers contain errors in the main theorems they claim to prove? Using GPT-6.1 Sol to identify such mistakes, it finds: - 6% of papers before 08/2026 contain main theorems that are false. - Only 1.5% contain similar mistakes in the last two months.
19
74
16
562
93,529
Results for Firstproof, Third Batch are out! Our agent, ProofCouncil, solves 9 problems :o
8
9
1
56
4,447
I had a bit of fun with ProofCouncil and managed to improve the bound in OpenAIs proof of the Quasi-Riemann Hypothesis a massive 4e-5, outperforming two other improvements already made since Tuesday by 5e-8. At this pace, we will solve the Riemann Hypothesis by 2109! 😅
2
5
2
48
7,087
As I have noticed that other people have found similar small improvements, I created this Website to track Lean verified progress on the constant: gtimg.github.io/QuasiRiemann… Everyone is invited to submit a Lean proof for a better constant, let's see how low we can get this!
1
1
14
1,622
The first person to reach constant 1/2 is hereby cordially invited to a hot or cold beverage of their choice in Zurich. :D
1
13
951
GPT-6.1 Sol is the new #1 on MathArena! Cheaper and better than Astra.
21
52
3
517
27,893
We just added GPT-6 Sol to MathArena! It comes in only just behind GPT-6 Astra, at slightly lower cost. Despite its significantly lower cost per token, it reasons more, bringing its cost closer to Astra.
3
9
80
3,240
We just added DeepSeek-v4.1-Flash on MathArena! Not quite on the level of Qwen-3.8, but boy is it cheap.
8
4
3
117
18,335
We just added Fable 5.1 (low) and Astra (low) to MathArena! At low reasoning, both still have an impressive performance, but Astra remains more efficient and better.
3
6
77
3,240
Unfortunately, we had to remove this entry: the authors ran this as an autoformalization task with access to source papers. Our pipeline run the models without such access. 80% is still impressive for open models on this challenging task, just not comparable to the other scores
Jasper Dekoninck
@j_dekoninckSep 11
We just added a new entry to ArXivLean! The Mistral team ran an agent consisting of two open models (Leanstral 1.5 as prover, Kimi K3 as coordinator) and obtained a massive 82% on the leaderboard, using an equally massive 10 million output tokens per problem.
6
8
1
124
12,231
Load more