No one wants to pay API pricing for AI usage. I put together a table of what subscription pricing value vs. low cost api value looks like across all of the most common models as of today, Nov 16th, 2026:
Key Takeaways:
Subscription value: sampled, cache-heavy workload, Claude Pro’s $20 plan extrapolated to about $706/month at API list prices, and Ultra’s $200 plan to about $6,829. Looks to be DOUBLE what @Openai is offering right now. By the way, if you use these tools, you know this feels pretty accurate.
Near the top in capability: Claude Opus 5.5 scored 59 on Artificial Analysis’s index, just behind Gemini Carbon at 60 and ahead of Astra at 57 in the tested modes. It looks stout. Its measured benchmark-task cost was higher
Strong capability per API dollar: GPT-6.1 Sol scored 55 at $1.27 per benchmark task, versus Astra’s 57 at $6.24. That’s a useful cost comparison, not proof the models perform identically.
Low-cost alternatives: DeepSeek Flash and MiniMax Flash were much cheaper per benchmark task, but also scored lower on that index.
My actual judgment: If your priority is maximum quality and extracting value from a heavily used subscription, @AnthropicAI deserves to be highlighted. Sorry @thsottiaux @ajambrosino @matteing
Would love to know more about Muse, Kimi if anyone has a way to get metrics!
Andrew retweeted
I can't believe I'm going to say this. I... actually like Gemini Neon. I think it's on par with Opus 5.5 if not better.
Andrew retweeted
Google AI Pro plan is going for like $6 for the first 3 months of your subscription. I urge you guys to take it.
Something kind of huge is happening that I'm starting to take notice of. Using Antigravity, the 20.00 subscription goes a LONG way. On top of that, it's VERY good at computer use. I don't know if i would say its better than Codex, but it might be. Would love to see some stats re; that.
Computer use was the biggest reason I was so attached with Codex / @OpenAI . This truly might be the thing that makes me cancel my subscription. The limits seem to be on par with the generosity of Opus 5.5 as I'm using Neon. And these models are smartttttttttttt!
31
Andrew retweeted
According to Artificial Analysis, Opus 5.5 is no longer on the pareto line (best value per intelligence)
Love AA and all, but they really need some behavioral benches to smooth out this gap. Opus 5.5 is still the best value available for real long running code work.
are you KIDDING ME on these metrics?!
Absolutely astonishing if even REMOTELY close to accurate. @thsottiaux @ajambrosino you COOKED LIKE CRAZY
122
Andrew retweeted
OpenAI's flagship, GPT-6 Astra, is sitting 7th on Agent Arena right now.
Google's Gemini 4 Hydrogen launched today and is already 5th. And Hydrogen isn't even Google's top model.
Claude Fable 5.2 is still #1 with more than double Astra's score. GPT-6.2 has a long way to go 💀
anthropic: introducing our most powerful model yet, fable 5.2
google: introducing gemini 4 carbon and our latest fast model, hydrogen
meta: introducing muse ember, our latest AI model
openai: introducing... business profiles in chatgpt and the retirement of two voice modes
Andrew retweeted
Atheism absolutely requires miracles:
It needs a universe from nothing.
Order from chaos.
Mind from matter.
Morality from preference.
Meaning from meaninglessness.
So yes atheism is a position that requires faith.
Andrew retweeted
Very excited to release Gemini 4 Hydrogen! Incredible speed and intelligence for everyday coding, agentic work and all-round use. I think you'll love this model. More info in the blog - enjoy! blog.google/innovation-and-a…
Reminder to myself: Have you been using one tool almost exclusively for the past month whether it be Codex, Claude, MausBot, Hermes, Openclaw, etc? STOP! You're silo'ing yourself. Have you been using one model nearly exclusively because its been at least mostly efficient for you? STOP! There is likely something better. Have you been scrolling on twitter relentlessly trying to find the best practices? STOP AND USE YOUR MIND. Have you been switching tools every day? Stop and see how it performs for a week or two. When is the last time you checked the available plugins, skills, githubs, open source models, codex skills, sites, etc? Have you ignored good feedback that continues to come up through posts? Have you considered what you are doing that you should be offloading to an ai agent? Are there things you have offloaded that need to be picked back up again to see if there might be a better or less costly way to do them?
If you have an idea, stick with it for months, not days, if its really worth it. don't just hand it to an agent to spiral in the wrong direction, don't give up on it until its at its best.
264





















