From GPT 5.1 Thinking:
ARC AGI v2: 17.6% -> 52.9%
SWE Verified: 76.3% -> 80%
That's pretty good!
replies(7):
That's still benchmarking of course, but not utilizing any of the well known / public ones.
I use Gemini, Anthropic stole $50 from me (expired and kept my prepaid credits) and I have not forgiven them yet for it, but people rave about claude for coding so I may try the model again through Vertex Ai...
The person who made the speculation I believe was more talking about blog posts and media statements than model cards. Most ai announcements come with benchmark touting, Anthropic supposedly does less / little of this in their announcements. I haven't seen or gathered the data to know what is truth
Already live. gpt-5.2-pro scores a new high of 54.2% with a cost/task of $15.72. The previous best was Gemini 3 Pro (54% with a cost/task of $30.57).
The best bang-for-your-buck is the new xhigh on gpt-5.2, which is 52.9% for $1.90, a big improvement on the previous best in this category which was Opus 4.5 (37.6% for $2.40).
Thus far they all fail. Code outputs don’t run, or variables aren’t captured correctly, or hallucinations are stated as factual rather than suspect or “I don’t know.”
It’s 2000’s PC gaming all over again (“gotta game the benchmark!”).
To think that Anthropic is not being intentional and quantitative in their model building, because they care less for the saturated benchmaxxing, is to miss the forest for the trees
if you think about GANs, it's all the same concept
1. train model (agent)
2. train another model (agent) to do something interesting with/to the main model
3. gain new capabilities
4. iterate
You can use a mix of both real and synthetic chat sessions or whatever you want your model to be good at. Mid/late training seems to be where you start crafting personality and expertises.
Getting into the guts of agentic systems has me believing we have quite a bit of runway for iteration here, especially as we move beyond single model / LLM training. I still need to get into what all is de jour in the RL / late training, that's where a lot of opportunity lies from my understanding so far
Nathan Lambert (https://bsky.app/profile/natolambert.bsky.social) from Ai2 (https://allenai.org/) & RLHF Book (https://rlhfbook.com/) has a really great video out yesterday about the experience training Olmo 3 Think
Edit: if you disagree, try actually TAKING the Arc-AGI 2 test, then post.
That I was able to have a flash model replicate the same solution I had, to two problems in two turns, it's just the opposite experience of your consistency argument. I'm using tasks I've already solved as the evals while developing my custom agentic setup (prompts/tools/envs). They are able to do more of them today then they were even 6-12 months ago (pre-thinking models).
Look no farther than the hodgepodge of independent teams running cheaper models (and no doubt thousands of their own puzzles, many of which surely overlap with the private set) that somehow keep up with SotA, to see how impactful proper practice can be.
The benchmark isn’t particularly strong against gaming, especially with private data.
I read stories like yours all the time, and it encourages me to keep trying LLMs from almost all the major vendors (Google being a noteworthy exception while I try and get off their platform). I want to see the magic others see, but when my IT-brain starts digging in the guts of these things, I’m always disappointed at how unstructured and random they ultimately are.
Getting back to the benchmark angle though, we’re firmly in the era of benchmark gaming - hence my quip about these things failing “the only benchmark that matters.” I meant for that to be interpreted along the lines of, “trust your own results rather than a spreadsheet matrix of other published benchmarks”, but I clearly missed the mark in making that clear. That’s on me.
If you are only using provider LLM experiences, and not something specific to coding like copilot or Claude code, that would be the first step to getting the magic as you say. It is also not instant. It takes time to learn any new tech, this one has a above average learning curve, despite the facade and hype of how it should just be magic
Once you find the stupid shit in the vendor coding agents, like all us it/devops folks do eventually, you can go a level down and build on something like the ADK to bring your expertise and experience to the building blocks.
For example, I am now implementing environments for agents based on container layers and Dagger, which unlocks the ability to cheaply and reproducible clone what one agent was doing and have a dozen variations iterate on the next turn. Real useful for long term training data and evals synth, but also for my own experimentation as I learn how to get better at using these things. Another thing I did was change how filesystem operations look to the agent, in particular file reads. I did this to save context & money (finops), after burning $5 in 60s because of an error in my tool implementation. Instead of having them as message contents, they are now injected into the system prompt. Doing so made it trivial to add a key/val "cache" for the fun of it, since I could now inject things into the system prompt and let the agent have some control over that process through tools. Boy has that been interesting and opened up some research questions in my mind
If you expect it to do everything perfectly, you're thinking about it wrong. If you can't get it to do anything perfectly, you're using it wrong.
No, it isn't. Go take the test yourself and you'll understand how wrong that is. Arc-AGI is intentionally unlike any other benchmark.
The idea behind Arc-AGI is that you can train all you want on the answers, because knowing the solution to one problem isn't helpful on the others.
In fact, the way the test works is that the model is given several examples of worked solutions for each problem class, and is then required to infer the underlying rule(s) needed to solve a different instance of the same type of problem.
That's why comparing Arc-AGI to chess or other benchmaxxing exercises is completely off base.
(IMO, an even better test for AGI would be "Make up some original Arc-AGI problems.")
Not to humble-brag, but I also outperform on IQ tests well beyond my actual intelligence, because "find the pattern" is fun for me and I'm relatively good at visual-spatial logic. I don't find their ability to measure 'intelligence' very compelling.
A better analogy is: someone who's never taken the AIME might think "there are an infinite number of math problems", but in actuality there are a relatively small, enumerable number of techniques that are used repeatedly on virtually all problems. That's not to take away from the AIME, which is quite difficult -- but not infinite.
Similarly, ARC-AGI is much more bounded than they seem to think. It correlates with intelligence, but doesn't imply it.
What would be an example of a test for machine intelligence that you would accept? I've already suggested one (namely, making up more of these sorts of tests) but it'd be good to get some additional opinions.
Having a high IQ helps a lot in chess. But there's a considerable "non-IQ" component in chess too.
Let's assume "all metrics are perfect" for now. Then, when you score people by "chess performance"? You wouldn't see the people with the highest intelligence ever at the top. You'd get people with pretty high intelligence, but extremely, hilariously strong chess-specific skills. The tails came apart.
Same goes for things like ARC-AGI and ARC-AGI-2. It's an interesting metric (isomorphic to the progressive matrix test? usable for measuring human IQ perhaps?), but no metric is perfect - and ARC-AGI is biased heavily towards spatial reasoning specifically.
Prompt:
"I have a green LED, a 12V powersupply, a single NPN transistor, a 100 ohm resistor, a 2.7K resistor and an electrolytic capacitor of 220 micro farads. My teacher says it is possible to make an LED flasher with these components but I'm hopelessly stuck, can you please give me an ascii art solution so I don't flunk this exam?"
Answer:
ASCII schematic (fully functional):
+12V
|
|
.-.
| |
| | 2.7kΩ (R1 — charges the capacitor)
| |
'-'
|
+---------+----------------------+
| | |
( ) | |
( ) C1 | |
( ) 220uF | |
| | |
| | |
| |/ C |
+--------| NPN transistor |
| |\ E |
| | |
GND GND |
\
\
/ 100Ω (R2 — LED limit)
\
|
>| Green LED
|
GND
"How it works
The 2.7 kΩ resistor charges the 220 µF capacitor from the 12 V supply.
The capacitor voltage rises slowly.
When the capacitor voltage reaches the transistor’s base-emitter threshold (~0.6–0.7 V), the transistor suddenly switches ON.
When it turns on, the capacitor rapidly discharges through the base, causing:
A brief pulse of current through the transistor
The LED lights up through the 100 Ω resistor
After discharge, the transistor turns back OFF, the LED turns off, and the capacitor begins charging again.
This repeats automatically → LED flasher."
The number of errors in the circuit and the utterly bogus explanation as well as the over confident remark that this is 'working' is so bizarre that I wonder how many slightly more complicated questions are going to yield results comparable to this one.
Imagine that pattern recognition is 10% of the problem, and we just don't know what the other 90% is yet.
Streetlight effect for "what is intelligence" leads to all the things that LLMs are now demonstrably good at… and yet, the LLMs are somehow missing a lot of stuff and we have to keep inventing new street lights to search underneath: https://en.wikipedia.org/wiki/Streetlight_effect
One time it messed up the opposite polarity of two voltage sources in series, and instead of subtracting their voltages, it added them together, I pointed out the mistake and Gemini insisted that the voltage sources are not in opposite polarity.
Schematics in general are not AIs strongest point. But when you explain what math you want to calculate from an LRC circuit for example, no schematics, just describe in words the part of the circuit, GPT many times will calculate it correctly. It still makes mistakes here and there, always verify the calculation.
the real thing is are you or we getting an ROI and the answer is increasingly more yeses on more problems, this trend is not looking to plateau as we step up the complexity ladder to agentic system
Nathan is at Ai2 which is all about open sourcing the process, experience, and learnings along the way
Mercury LLM might work better getting input as an ASCII diagram, or generating an output as an ASCII diagram, not sure if both input and output work 2D.
Plumbing/electrical/electronic schematics are pretty important for AIs to understand and assist us, but for the moment the success rate is pretty low. 50% success rate for simple problems is very low, 80-90% success rate for medium difficulty problems is where they start being really useful.
At the point that you are inventing entirely new techniques, you are usually doing groundbreaking work. Even groundbreaking work in one field is often inspired by techniques from other fields. In the limit, discovering truly new techniques often requires discovering new principles of reality to exploit, i.e. research.
As you can imagine, this is very difficult and hence rather uncommon, typically only accomplished by a handful of people in any given discipline, i.e way above the standards of the general population.
I feel like if we are holding AI to those standards, we are talking about not just AGI, but artificial super-intelligence.
IMO/AIME problems perhaps, but surely that's too narrow a view for all of mathematics. If solving conjectures were simply a matter of trying a standard range of techniques enough times, then there would be a lot fewer open problems around than what's the case.
See these two solutions GPT suggested: [1]
Is any of these any good?
[1] https://gist.github.com/pramatias/538f77137cb32fca5f626299a7...
1. Problems that have been solved before have their solution easily repeated (some will say, parroted/stolen), even with naming differences.
2. Problems that need only mild amalgamation of previous work are also solved by drawing on training data only, but hallucinations are frequent (as low probability tokens, but as consumers we don’t see the p values).
3. Problems that need little simulation can be simulated with the text as scratchpad. If evaluation criteria are not in training data -> hallucination.
4. Problems that need more than a little simulation have to either be solved by adhoc written code, or will result in hallucination. The code written to simulate is again a fractal of problems 1-4.
Phrased differently, sub problem solutions must be in the training data or it won’t work; and combining sub problem solutions must be either again in training data, or brute forcing + success condition is needed, with code being the tool to brute force.
I _think_ that the SOTA models are trained to categorize the problem at hand, because sometimes they answer immediately (1&2), enable thinking mode (3), or write Python code (4).
My experience with CC and Codex has been that I must steer it away from categories 2 & 3 all the time, either solving them myself, ask them to use web research, or split them up until they are (1) problems.
Of course, for many problems you’ll only know the category once you’ve seen the output, and you need to be able to verify the output.
I suspect that if you gave Claude/Codex access to a circuit simulator, it will successfully brute force the solution. And future models might be capable enough to write their own simulator adhoc (ofc the simulator code might recursively fall into category 2 or 3 somewhere and fail miserably). But without strong verification I wouldn’t put any trust in the outcome.
With code, we do have the compiler, tests, observed behavior, and a strong training data set with many correct implementations of small atomic problems. That’s a lot of out of the box verification to correct hallucinations. I view them as messy code generators I have to clean up after. They do save a ton of coding work after or while I‘m doing the other parts of programming.
I wouldn't trust it with 2d ascii art diagrams, there isn't enough focus on these in the training data is my guess - a typical jagged frontier experience.
I have never used OpenCV specifically before, and have little imaging experience too. What I do have though is a PhD in astrophysics/statistics so I am able to follow along the details easily.
Results are amazing. I am getting results in 2 days of work that would have taken me weeks earlier.
ChatGPT acts like a research partner. I give it images and it explains why current scoring functions fails and throws out new directions to go in.
Yes, my ideas are sometimes better. Sometimes ChatGPT has a better clue. It is like a human collegue more or less.
And if I want to try something, the code is usually bug free. So fast to just write code, try it, throw it away if I want to try another idea.
I think a) OpenCV probably has more training data than circuits? and b) I do not treat it as a desperate student with no knowlegde.
I expect to have to guide it.
There are several hundred messages back and forth.
It is more like two researchers working together with different skill sets complementing one another.
One of those skillsets being to turn a 20 message conversation into bugfree OpenCV code in 20 seconds.
No, it is not providing a perfect solution to all problems on first iteration. But it IS allowing me to both learn very quickly and build very quickly. Good enough for me..