Poetiq achieves SOTA on the Arc-AGI-2 semi-private eval.

They do it with harnesses on frontier models rather than RL or new models. This task-specific optimization/specialization is The Way.

poetiq.ai/posts/arcagi...

#AGI benchmarks should be developed by neutral orgs completely in private with no contact to the internet. Candidate #LLM would then be tested. This makes it impossible to train or fine-tune models on any benchmarks. Only afterwards results would be published. The only problem: Leaking #ARCAGI #ABAP

@melaniemitchell.bsky.social’s article sheds light on a genuine breakthrough in #AI, a shift that redefines its limits. Are we edging closer to human-level reasoning in ARC-AGI? If so, it’s a game-changer, and our understanding of AI will need a serious update. #ARCAGI #AIBenchmark #OpenAI

poetiq.ai/posts/arcagi...

Poetiq's methodology on top of Gemini 3 & GPT 5.1 exceeds average human performance on ARC-AGI-2!
This is huge.

Only caveat is that they evaluated on the public set - it might have been used in post training of Gemini 3? Looking forward to see private eval results! poetiq.ai/posts/arcagi...

Eso siempre, pero lm arena es el benchmark más inútil. Yo quiero ver livebech o arcagi

For smaller grids on the #ARCAGI test you may call #o3 "superhuman" (this depends on how you define superhuman). For larger grids the performance falls very quickly to below human performance.

This may be directly related to the amount of tokens involved as grid size increases.

🔥 #OpenAI o3 model performance makes a leap, sets a new high score on the #ARCAGI benchmark.

Source: arcprize.org/blog/oai-o3-...

#ml #ai #arcagi #benchmark #openai

Read the full analysis and get our code: poetiq.ai/posts/arcagi....

The new Tiny Recursive Model (TRM) uses a two‑layer network with just 7 M parameters and reaches 45 % accuracy on ARC‑AGI‑1 and 8 % on ARC‑AGI‑2, outperforming larger LLMs. Read more: https://getnews.me/tiny-recursive-model-beats-large-models-on-arc-agi-puzzles/ #tinysrecursivemodel #arcagi #llm

OpenAI、次世代AIモデル「o3」を発表、ARC-AGIテストで”85%超え”の快挙達成
#o3 #12DaysofOpenAI #o3mini #ARCAGI #ITニュース

poetiq.ai/posts/arcagi...

If AI flunks François Chollet’s test, maybe it just struggles with colorful grids—not intelligence itself.
#AI #AGI #Intelligence #Chollet #ARCAGI #PhilosophyOfAI

So I suppose the best thing to do today is to stare at some o3 output data on ARC-AGI. Here's a simple visualization on the public eval with o3's attempts and gt solutions. (Pls don't spam)
arcagi-o3-viz.netlify.app

Yay! Shit sandwiches are getting cheaper!

poetiq.ai/posts/arcagi...

DeepSeekの推論モデル「DeepSeek-R1」をOpenAIのo1＆o3と比較することで明らかになったこととは？
#DeepSeekR1 #ARCAGI #ARCPrize #ITニュース