alt Hacker NewsGPT 6.1 Sol: Near-Astra intelligence for a fifth of the price
Comments
Really good Hamster[0], and in my tests it's very close to Astra, and indeed ~5x cheaper.
Actually, it is so close to Astra, that I'm starting to beleive it's a slightly quantized version of Astra.
[0]: https://aibenchy.com/compare/openai-gpt-6-1-sol-xhigh/openai...
There was a model called Astra-Minor, found in the files a few days ago. I assume Sol 6.1 is this, as a last minute panic rename due to Sol 6 being underwhelming while Opus 5.5 turned out really strong. I can't really explain releasing Sol 6 in any other way, especially mere days ago.
> Cached input costs just $0.10 per million tokens—95% less than standard input pricing and 50% less than GPT‑6 Sol’s cached input pricing
This is the actual big announcement. 50% cheaper cache than GPT-6 Sol will get you far more mileage on Codex.
The GPT 6 release was ... not great.
Sol 6 was so bad that I switched over to Opus 5.5 exclusively.
Huge regression compared to Sol 5.6, often doing really dumb things. Same for Luna.
Even Astra is very unreliable for coding. Brilliant for vision, sometimes just great, but it also often does very stupid things.
I'm a bit sour on OpenAI right now and skeptical that 6.1 will be much different.
(Note: this is after preferring and shilling Codex/OpenAI models for the last half year)
Ominous for the industry and investors that token price is becoming the main battleground. Could be Anthropic's rationale for IPOing this year.
I must say that this AI thing is going more or less as I felt it would back about a year ago. I think there is no real moat in AI models. It's a commodity and the big labs have predictably been caught in a race to the bottom. Not sure if this is going to turn better or worse for all of us common folks. I must say I'm a bit happy though in the sense that "intelligence" is not going to be controlled and be rented out by a small minority.
I'm a bit late with the pelicans because I was live-blogging the keynote: https://simonwillison.net/2026/Sep/29/openai-devday-2026-liv...
Here they are for GPT-6.1-Sol: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...
They're not notably different from the GPT-6 family pelicans: https://static.simonwillison.net/static/2026/gpt-pelicans-gr...
I love free market competition. We're getting insane advancements every day. I remember when llms used to cost an arm and a leg for decent intelligence
Can someone explain to me why on these benchmarks like these a higher effort level often has a lower score?
For example, GPT-6.1 Sol High gets 75.2% on DeepSWE and XHigh gets 71.9% and is more expensive
https://openai.com/index/introducing-gpt-6-1-sol/#deepswe
Also, how many times did they test each condition - just once or a few times? are they showing an average of multiple attempts, etc..
Astra requires multiple turns and fresh refactoring agents to produce good code.
Fable 5.1/Opus 5.5 isn’t different, but the first cut is better quality.
Astra is a whole order of magnitude cheaper than Fable, and the Anthropic usage limits are ridiculous. Layers on layers of limits that constantly trip.
We don’t really use Sol because Astra X High is cheap. Some have mentioned regressions but we haven’t noticed any with Astra.
What's driving the increase in release cadence here? We seem to get new models every week or so now, is this RSI?
“OpenAI's new Pro 500 plan offers OpenAI's highest usage allowance and comes with access to its new "Ultrafast" feature — it also costs $500 per month.
At the same time, OpenAI is also making its existing $200 Pro plan less appealing. In Codex and Work, $200 Pro subscribers will see their included usage decrease from 20x of what the company offers to Plus users, down to 10x of that same allowance. In ChatGPT, meanwhile, GPT-6 Pro message caps will decrease from 200 to 100 per week.”
https://www.engadget.com/2272106/openai-adds-dollar500-pro-s...
Yikes
Looking at the token prices, if this is half as good as 6-Astra for 3D model creation in Blender, it's going to be an absolute game changer.
Opus 5.5 is definitely better at coding, but nothing even comes close to 6-Astra for work in 3D graphics...
6.0 Sol was literally a week ago... Basically continuous integration for model releases at this point.
Since Luna is so dirt cheap compared to Sol/Astra it would be nice if they could set or you could reserve some small percent like 3-5% of usage pool on codex just for Luna so if you hit usage limits you can at least still run a lot of Luna.
> GPT‑6.1 Sol matches GPT‑6 Astra at roughly one-fifth of the cost
Astra is a pretty impressive model. Excited to try this.
Opus 5.5 is on another level, especially when it comes to mathematics implementations. You can drop it a PhD-level physical simulation (for example, a contrast-injection simulation for angiography in my case), and it just...implements it. With full-on WebGL rendering in the browser, from scratch (or using an existing library, if you prefer).
If the Terminal Bench 4.0 scores are to be believed[0] GPT-6.1 is an incredibly efficient model.
Yes, benchmarks aren't real work blah blah, but the delta here is so large compared to Astra, it makes it seem like this is distilled Bel or similar.
GPT 6 Sol is obsolete after only one week! I am glad that they are not afraid to update the models more frequently. The Navier-Stokes thing revealed that it took them only a week or two to train a model more capable than Astra, and I want the pace of public releases to keep up with that.
I find GPT 6 to be lacking in common sense when it comes to interpreting my prompts.
I have to be more literal with it than with GPT 5.x, otherwise, it sometimes does something totally different than what I want.
Impressive improvements, but GPT 6 Sol came out 7 days ago, and this one will behave differently. The panicked pace is becoming a liability, maybe they should have waited and released this as the 6.0 release
They released GPT 6 Sol literally 6 days ago. We've accelerated to a weekly model release cadence. That seems like...a big deal.
OpenAI is feeling really competitive again.
I just added an agent / coding agent into an email app, and doing it through `codex` and its Codex App Server couldn't have been easier, and the results are very compelling.
The open source harness, API around it, and friendliness for connecting a subscription puts Claude to shame right now.
A few more thoughts here https://housecat.com/blog/introducing-housecat-agent
Here's a comparison of a image->html flow for GPT 6.1 Sol vs Opus 5.5.
GPT 6.1 Sol: https://html.non.io/lcars-gpt-6.1-sol
Opus 5.5: https://html.non.io/lcars-opus-5.5
Overall, opus executes a bit better than 6.1 sol, which surprises me. Astra has been the best model for this flow so far, so the fact that Sol missed some alignment / vision pieces here is interesting. It's not bad by any means, but I think where Opus really wins is the motion animation of the svgs / final polish (scroll down to the "customize every detail" section on the homepage, the svg animation is beautiful for that).
Still, it executed quick and was quite cheap to run.
For most sane people, OpenAI is the way to go... A lot of usage with very good models, but you know that Anthropic is laughing all the way to the bank with Opus 5.5 being "the best" model right now... There are a ton of people (and companies) that will just refuse to use anything else than the highest benchmarking model in existence.
It seems pretty clear that this is a much larger model than Sol 6, and you can see this in the much lower generation times. I think this is also the main explanation for the $200 plan being cut in terms of API usage.
This is because they have really aggressively priced a larger model to compete with Opus 5.5, so their margins are much worse. Consequently, the equivalent API spend on the subscription is much less.
I wonder if releasing this soon sort of validates the rumor that Sol 6 was just the Terra model they bumped up and slashed the price.
Then Opus 5.5 caught them off guard and now they're actually releasing the correct sized model.
This is great. But maybe part of the motivation is that 6-Sol wasn't as good as initially advertised so they needed to tweak it. I felt a clear degradation in quality in some simple refactoring tasks vs 5.6-Sol.
Frontier models are being used to obtain training data from users. We burn tokens teaching OpenAI how to make a cheaper model that is almost as good. I think the new $500/month pricing strategy is a significant misstep by someone who has clearly not tried Gemini 3.8 Flash or Deepseek 4.1 Flash.
sol-6 is terra-6. They figure that no one was using terra and they could bring the speed and cost saving of terra distilled on astra, but rebranded as the more popular sol.
Back fired because of opus 5.5.
So now we get the real sol-6 as sol-6.1, and OpenAI will eat the cost to stay competitive.
This could be invalidated if sol-6.1 is the same speed as sol-6.
I got a popup in my Codex just now saying "Try out 6.1 Sol!" and so I clicked the button to try it, and intriguingly, it set my model selector to "GPT-6 Astra Light" which makes me think 6.1 Sol may be in some way just a lighter/distilled version of Astra? defo interesting, not sure if I should read too much into it though. I see no option for directly selecting 6.1 Sol in my Codex Desktop UI.
> Not sure what kind of usage can justify $200/month of either openai or anthropic, i'm not even talking about $500
It's easy to hit those numbers in a day in an modern-enterprise context synthesizing from incoherent information in jira, slack, layers of codebases etc. Modern enterprise meaning a firm that has been serving a few strategic customers w/ "move fast and break things" since day 1
Just got access in Codex, looking forward to trying it out. Opus 5.5 has blown me away with what it's capable of doing, hopefully 6.1 will actually be a worthwhile contender.
Excited to tryout Decisions API as well.
Do these benchmarks have any meaning anymore? And do the announcements seem less exciting now? (Not taking anything away from the advances we are making but it seems more incremental now?) The reliable way to tell if you'll like a model is reliable collage/X reviews to gauge a model's capability and then trying it out to see if you like the style.
The last time a model announcement felt like a leap in capability beyond other things out there was Fable - which was promptly taken away. Sol and recently Opus 5.5 were strong because they approach that capability with a lot more efficiency and don't blabber incoherently (looking at you Opus 5.1).
Deepseek is a workhorse for those who prefer open and API usage. Other than that the model announcements all just seem like a blur and quite interchangeable but I wonder if that's just me tuning out or do others feel the same way?
Price/intelligence comparison with Opus 5.5 on Artificial Analysis:
https://artificialanalysis.ai/?models=gpt-5-6-luna-low%2Ccla...
According to this, at Max it's better and cheaper than 5.5 Medium, but worse than 5.5 High. At Medium, it's better and cheaper than 5.5 Low.
Surprisingly (or maybe not) it matches the performance of Astra on my benchmark[1], but is much cheaper. It is also head to head with Opus 5.5 on both the price and pass rate, but edges it out slightly.
Hardly any comparisons to Opus 5.5, which means it's not great
The real announcement is the ultra fast mode ... Astra at 300t/s is insane!
Ok, but we need 6.1 Luna soon. 6 feels worse than 5.6 in our agentic use case.
Is it really 1/5 of the price if most people who use it are also losing 1/2 of their credits?
I feel a little salty about the plan changes. I wanted to upgrade to the $200 plan a day after it was blocked. Now it only includes half the usage unless for those that got grandfathered into the x20 usage.
What happened to "we urge you to urge us to stop moving AI so fast"?
Wasn't 6 released like last week? I can't keep up anymore.
The main issue I have is how they nerf their models and the quality difference between API users and their subscribers.
API price cuts were obvious once they made their announcement changing how usage is counted.
These moves all make sense when you take into account the enterprise market.
It’ll be interesting to see what happens to the economics of this business if we hit a wall on peak intelligence but keep finding cool ways to lower prices.
Cache is priced at $0.1/M, 50% as sol 6 and sonnet 5.5.
I am yet to spend $200 on deepseek this year. Not sure what kind of usage can justify $200/month of either openai or anthropic, i'm not even talking about $500. Deepseek is faster, IMO intelligence difference is negligible and it so much cheaper that i no longer care about how much i use it. I never hit any daily/weekly quota or anything like that while working or tinkering. At this point i am OK with being 6 months behind the "frontier", purely on bang-for-buck basis and who cares which shadowy government gets my data.