GPT-6 Sol and Luna

At this point, the deciding factors for me between Claude Code 20x and Codex Pro 20x are:

1/ Usage limits: downstream of input/output cost, but resets and obscure windows and odd 20x plan / 5x plan != 4x usage math throw a wrench into it. Winner right now is Codex by a mile, especially when you factor in ChatGPT usage (even 6 Astra Pro) is essentially unmetered on the 20x plan. Always a bummer when asking if I should see a doctor about a rash means I can't code as much. It's also is a godsend if you use an MCP like oracle to automate the process of calling the Pro model on particularly tough problems, giving better planning results or deeper code analysis without burning usage.

2/ Context window in the harness. Claude Code wins on this. There used to be a toml file workaround for Codex to extend the GPT context window to 1m, but this stopped working on the plans and only on per-token billing. 252k is just not enough. Codex's compaction is very good, fwiw, but it happens so frequently that even a model as powerful as Astra sometimes loses the plot on long-running tasks.

3/ Ability to use the plan outside of the official harness. Codex wins. Anthropic does shit like bills requests as extra usage if it sees a hermes.md in a commit.

I've subscription hopped a bunch, and at times I've had both, but I keep coming back to Codex because it wins on 2/3.

GPT-6 Luna being half the price of GPT-5.6 Luna is a really big deal.

Here's GPT-6 Luna pelicans: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

And GPT-6 Sol: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

Scroll to the bottom for the GPT-6 Sol max one: https://tools.simonwillison.net/markdown-svg-renderer?url=ht...

For comparison, here are the pelicans I got for GPT-6 Astra: https://tools.simonwillison.net/markdown-svg-renderer?url=ht... - I still like the Astra Max one best.

I've been working with agents all year, but 5.6 Sol was some sort of sweet spot for me. Something about how it communicated verbally and its engineering instincts just clicked for me, and I was able to somehow predict it and jam with it. Like a colleague you click with. It's the first model I've gotten attached to. I'm concerned that whatever model supercedes it, while technically better, just won't feel quite as natural to work with. And this makes me feel very professionally vulnerable to the labs. I miss the days when my crucial tooling came from companies as reliable and predictable as, say, Jetbrains.

I don't see how anyone can be using Claude with prices like this, it's pretty incredible what the OpenAI team is doing, w.r.t model quality and pricing.

  Prices per 1M tokens     Claude Opus 5.5    Claude Opus 5
   Cache reads              $0.20              $0.50
   Input tokens             $4                 $5
   Output tokens            $20                $25
   Cache writes             $5                 $6.25

Input

Output

Price reduction

GPT‑6 Sol vs. GPT‑5.6 Sol

$4 → $2

$20 → $10

50% cheaper

GPT‑6 Luna vs. GPT‑5.6 Luna

$0.20 → $0.10

$1.20 → $0.50

50% cheaper

Have they solved GPT5.6 SOL's propensity to over-engineer and over-complicate? You'd ask SOL to do something relatively simple, and find four single-use methods, an interface, and a factory-factory.

I actually preferred 5.6-Terra not because it is technically superior (it isn't) but because it had better instincts to NOT do this stuff.

PS - Speaking of better instincts, have they closed the UI-design gap at all? I keep a Claude subscription just because /design produces significantly higher quality UI design/UI feedback/UI refinement than anything I've seen from OpenAI.

I’m still finding context is king, even with the best models.

For example, I had Fable review Astra’s output yesterday, and it found some issues and fixed them. Passing the fixes back, Astra then uncovered additional issues with Fable’s fixes (and yes, this will go on ad infinitum if you let it, but these were “real” issues).

It seems the big story here is the reduced Luna pricing. It’s a fantastic model that can handle most automation needs (though I still use the big models for day-to-day development).

The only two benchmarks shared between the Opus 5.5 and Sol 6 launch seem to be frontier code and automation bench, looks like Sol wins on automation bench (same performance for half the cost) and Opus 5.5 wins on frontier code (2-5% better scores across the board for same cost)

Good. Maybe they can use GPT-6 to fix the accessibility of their iOS app. Output shows as text fields to VoiceOver, and the accessibility announcements have backslashes before seemingly every punctuation mark. And then bring accessibility announcements to the Android app so I don't have to make a whole new app just to add that through an accessibility service. Ugh the things I do for accessibility cause I'm blind. On a better note though, AI has done so much for the blind community, from image (and increasingly video) description to mods for video games like Final Fantasy 1 through 6 Pixel remaster, I have a ton to be grateful for.

Cutting prices by 50% as compared to 5.6 prices is exciting. GPT-6 Luna at $0.10/Mio input tokens and $0.50/Mio output is positively insane.

I'm glad that both labs noticed and are trying to improve the models communication styles, they were getting closer and closer to annoying gibberish.

Artificial Analysis is reporting that 6 Luna scores 2 points lower on their coding index than 5.6 Luna, but is 60% cheaper:

> In the Coding Agent Index, Sol improves but Luna regresses: In OpenAI's Codex harness, GPT-6 Sol (max) scores 57 in the Artificial Analysis Coding Agent Index, up 2 points from GPT-5.6 Sol (max), with gains in Terminal-Bench 4.0 (43% vs 37%) and SWE-Atlas-QnA (58% vs 54%). At $2.99 per task it costs ~50% less than GPT-5.6 Sol (max) and sits on the Pareto frontier of Coding Agent Index vs Cost per Task. GPT-6 Luna (max) scores 41, down 2 points from GPT-5.6 Luna (max), with lower scores in SWE-Atlas-QnA (44% vs 49%) and DeepSWE v1.1 (64% vs 66%), at ~60% lower cost per task.

https://x.com/ArtificialAnlys/status/2102462962758033624

Does anyone know if the ~50% price reduction also implies x2 subscription usage? Or is it only for the API.

Edit: Yes, it applies also to subscriptions, source https://x.com/thsottiaux/status/2102463847714247142

Not only is GPT-6 Luna better, it's 50% cheaper. It was already practically free on a pro plan.

OpenAI, Antrophic and others are operating with 80% margins. They can lower the prices for a long while.

It's a lot to ask to trust they can or will maintain this new pricing. In any case it's exciting to think this could push further price cuts in the highly competent and competitive Chinese clones. I'm still using ChatGPT for interactive queries, but at this point pretty much only because of its familiar UI

I've been raving about Luna 5.6 as it's dirt cheap, and "intelligent enough". Double quoted.

Now GPT 6 Luna is even cheaper, and more intelligent, there is no going back... to SOL 5.6 for intelligent layer.

Luna really is "intelligence to cheap to meter" by now

No terra it seems? Luna 5.6 is great for token churning so it will be exciting to try the new one.

Opus 5.5 seems better? Can someone attach both scores

Why is Claude missing on the "Factuality" graph?

Code deception 10% at 5.6 to 1.3% for 6.0? So models are getting more safe rather than less safe? hmmm

Live notification in Codex:

> GPT-5.6-Sol is retiring. This conversation will automatically switch to GPT-6-Sol

I don't recall OAI retiring a model so early lol. Similar arch?

Would be funny if Google drops Gemini 4 today.

6 Sol Performs worse than 5.6 Sol at DeepSwe?

Wierd!!

This Luna pricing is obscene man. 5.6 was good enough for so many use cases (data analysis, structured extraction etc).

Incredible.

Gpt 6 Luna is cheaper than Deepseek 4.1 flash! Today is wild in terms of intelligence/price across the board!

It looks like the optimal pattern is to have Astra as the orchestrator and Sol as the implementer. Same as with Fable and Opus.

I dont understand why there is not a gpt-6 terra?

Love the price reductions across major players

So opus 5.5 has reduced price. Who is winning then?

Pricing is insane, can have Luna going after a goal for 10 days and not run into maxing out the limits.

What a day! I couldn't really use the last Luna for much (wasn't smart enough) or Astra (too expensive). So this release is really exciting. I can probably use Sol 6 as much as I want in the week, which as great.

There's no way this wasn't meant to coincide with Anthropic's release today.

Feels like Opus 4.6 / GPT-5.3-Codex all over again: Anthropic gets a few hours to enjoy the launch, then OpenAI drops something the same day.

so opus 5.5 is smarter and cheaper than fable, and sol 6 is a little dumber and WAY cheaper than astra? is that right?

Selling dollar bills for $0.40 to undercut the guys selling them for $0.50 is a bold move. Let's see if it pays off for them.

Third-party inference providers will have a hard time to beat Luna in pricing with comparable open models.

Very nice in cost/1mtok. Looks like more work is being done for efficient everyday helper models as time goes on.

This is the most blatantly astroturfed thread I have ever seen on Hacker News.

My previous comment--which suggested astroturfing--was the highest upvoted comment here until it got flagged. Which implies to me that atleast the other remaining humans on this forum see it as well.

26 minutes, 89 comments, upvoted instantly to the top, posted within one hour of the Opus 5.5 announcement. You tell me.

Luna is the goat for real, cost intelligence ratio is insane already and it is enough for most daily computer use.

The new default is 6.0 Sol high. Escalate to Astra-medium. If usage is tight go luna6.0-max

Looking at their own charts it seems like it's only small incremental improvement over 5.6 Sol, but with a massive cost reduction. And the better writing/communication style that Astra had.

Which... fine, I'll take that.

Thirteen comments in six minutes, primarily one liners about how "insane" and "very nice" the new prices are, and how "Claude can't compete"? Within one hour of the Opus 5.5 launch announcement?

Goodbye Hacker News--it has been a good run.

sage

As a user of 5.6-Terra, I am sick and tired of the inconsistencies in GPT model families. There is no 6-Terra.

Well at least it looks like OpenAI is dogfooding because their announcements, product names, and everything else looks and sounds like LLM-slop.

[dead]

[dead]