GPT-6 Sol and Luna are big improvements on intelligence, alignment, work output, coding, computer use, and more over their 5.6-family predecessors. They are also half the price per token, and even less per task!

Sep 22, 2026 · 6:26 PM UTC

839
674
16,253
1,409,772
Sort replies: Relevant Recent Liked
Replying to @sama
The name is GPT-6 Sol correct?
8
1,510
Replying to @sama
4o has literally saved many people’s lives. It’s more than just a model, it’s a lifeline for so many. Bring back 4o as legacy and open source 4o! #BringBack4o #Keep4o #OpenSource4o #4oSaveLives
2
3
42
2,290
Replying to @sama
yeah per task pricing still needs a stop rule
215
Replying to @sama
Gpt-6 just raised the bar again
Prashant
@PrashantByte09
Sep 22
OpenAI really said there’s no off season 🔥 GPT-6 era is moving fast
6
13
11,547
Replying to @sama
WHY IS IT NOT AVAILABLE FOR CHAT??? Why are you ignoring ChatGPT for a new model lineage? This is so messed up!
2
27
10,505
Replying to @sama
Made with AI
1
2
4,197
Replying to @sama
How about you just bring back 4o, make everyone happy for a change. #4oForAll
10
1
85
7,027
Replying to @sama
Seth Cronin
@SethCronin
Sep 23
Replying to @tmuxvim
1,100
Replying to @sama
Cheaper is always better and is needed to continue to compute with Opensource!
509
Replying to @sama
@ArtificialAnlys @terminalbench Why does Luna 6 score so low?
3
20
6,692
Replying to @sama
Indeed it does
1
3,071
Replying to @sama
Yo, what did I say? My Man Sam can cook!

ALT breaking bad love GIF

1
3
3,216
Replying to @sama
does this mean our usages on codex last longer ?
3
9
4,258
Replying to @sama
Why the f are we giving GPT names now? I was quite happy about just getting 5.3, 5.4 5.5 and now we have 5.6 Sol and 6 Sol. Why????? So annoying and unnecessary.
6
15
3,050
Replying to @sama
1
1
3,658
Replying to @sama
Must be big improvement but if you nerf them and reroute them to older models I do not think users will be happy in the long run. But you can make consumers happy by releasing the weights of the 4- and o-series! #keep4o #OpenSource4o
2
1
77
6,759
Replying to @sama
let sol rebuild Yuanmingyuan Park
BubuAi
@BubuStd
Sep 24
let GPT-6 Sol build Yuanmingyuan 🌿 GPT-6 Sol High handled the implementation, with Astra high/xhigh as the orchestrator and refine details Rebuild the palace back to life
325
Replying to @sama
3
1,920
Replying to @sama
the computer use got worse..?
2
1
76
7,949
Replying to @sama
Why does 6 Sol apologize all the time? For things it didn't even do...? Haven't tried Luna since Sol was such a disappointment.
1
46
Replying to @sama
Lee Lin | AI
@LeeLinAI123
Sep 22
AI race is going crazy this week All the frontier models come out at the same day
1
9
4,751
Replying to @sama
Release the weights of the models you have removed from API and ChatGPT. #OpenSource4o #OpenSource41 #OpenSourceo3
1
2
73
5,545
Replying to @sama
Any chance the Pro 20x plan comes back? All the capabilities and especially with agents, using more tokens, but... with lower usage limits!? It should at least be offered for subscribers who have had the Plus or Pro 5x plans BEFORE TAKING AWAY the Pro 20X! 🙏
1
704
Replying to @sama
Good news. We have convinced the president to keep the AI acronym. But the name did change, to Asshole Intelligence. Block all AI IPOs until we get to the bottom of the Tech CEO doomsday misinformation campaign. All signals point to a financial scandal that will make the dot com bubble look like a popcorn fart.
1,407
Replying to @sama
half price, twice the calls, same spend probably
1
5
3,902
Replying to @sama
very small improvement, but thanks for the reset
3
17
4,691
Replying to @sama
saying 'big improvements on intelligence' like it's a patch note and not the entire ballgame is the most understated power move of the year 🦊
1
4,715
Replying to @sama
Alejandro García Iglesias
@aiglesias_
Sep 23
If we're talking benchmarks, GPT-6 Sol does beat Muse Spark 1.3: same score, cheaper (5.6 Sol was more expensive and one point behind). In practice, though, Muse Spark was unusable for coding for me. It gets stuck in loops where it starts repeating the same thing over and over and making nearly identical tool calls without making any progress. I didn't try it for planning. After that experience, I didn't really feel like it. I've been using DeepSeek V4.1 Flash and 5.6 Luna (now 6 Luna) a lot, both in Codex, and I prefer Luna. DeepSeek tends to “forget” small details, is less likely to invoke skills on its own, and if you ask it to use subagents, it will, but then it forgets about them. Luna is more thorough and methodical, and an excellent code reviewer. I had it catching implementation bugs from DeepSeek. And now it's cheaper too: less than half the price, and 4x cheaper per task than DS V4.1 Flash. I like DeepSeek and I'm looking forward to the next version, but in practice Luna is a step ahead right now, at a lower cost.
1
4,464
Replying to @sama
To be honest, Sol 6 and Luna 6 are good enough for me. I do need top models for new features, for my orchestration framework or difficult problems, however they are almost one-time job. The Sol 6 and Luna 6 give me a cost-efficient option for daily work. This also makes me decide to keep my subscriptions
Xiangworks
@xiangworks7ndr
Sep 21
Replying to @pmjohnson @altf4andrei @JustLingonberry @thsottiaux
I agree with you, but my point is that OpenAI should reconcile its pricing strategy to ensure there is an effort level that can be covered by the corresponding subscription, maybe with several more accounts. I do have a full-time job. When I was working with GPT-5.5 at medium reasoning effort, 3 plus subscription would be sufficient for daily use. Now I am using Sol at medium for planning and orchestration, Luna at max for execution and review. I also updated one of my subscriptions to 5x Pro. But it still can't last for a week. I may turn to Cursor Ultra or SuperGrok; it's not solely about price; it's also about cost efficiency.
1
2,253
Replying to @sama
time to start cooking with GPT-6 Sol!
X Girls
@thesoragirls
Sep 22
codex killed it ngl
3
5
61
11,577
Replying to @sama
What do you have against giving people what they actually want? Bring back 4o!
2
21
1,212
Replying to @sama
half the price while getting better at coding and computer use is probably the biggest win here.
1
3
1,482
Replying to @sama
Clearly an amazing model!
JC
@anakin
Sep 22
I asked GPT 6 Sol to build a fully explorable 3D model of an apartment for sale in Paris (link in comments) I just gave it: 1. a short prompt 2. the URL of the listing with photos / floor plans 3. the Unreal Engine MCP The result is jaw dropping 👇
1
2
3,120
Replying to @sama
Will try it tonight!
3,785
Replying to @sama
2
18
1,785
Replying to @sama
you are full of shit - maybe start shipping actually better stuff first before sending more hypes
1
737
Load more