Vano

Read the papers.

[Vano @vano_ai]

@sflorimm Lowest barrier for entry, fastest turnaround/instant gratification, and models are hyper optimized on web apps.

[Vano @vano_ai]

@josh_wills It’s interesting to see such stark differences in experience with opus 5 from people. For me it’s just slightly below Sol 5.6 performance but it gets the job done.

[Vano @vano_ai]

Correction, $75 with claude every 5 hours with the extra 50% usage so $1000 stock per week, $1500 during the promotion.

[Vano @vano_ai]

Update on further testing of the $20 OpenAI and Anthropic tiers in https://t.co/M7udj1x06A.

Opus 5 hour limit, $50 in credit usage, consumes 5% of weekly.

Sol 5.6 WEEKLY limit $100 total available in credit usage.

That translates to:
$1000 of subsidized usage with claude ($500 without the 2x usage promotion) v.s. $400 codex usage.

[Vano @vano_ai]

@ashwinvel94 When I made this post they advertised that doing deep learning research like creating new architectures would get flagged by the classifier along with cyber and other fields and refuse to work.

[Vano @vano_ai]

Everyone comparing Fable 5 vs GPT 5.6 Sol from the preview is doing the same as comparing Opus and Fable. No one has tried the Fable level GPT model (Ultra or Pro), why are we comparing the Opus competitor. Not to mention, gpt-5.6-sol is HALF the API cost of Fable... $5/30 per 1M, vs $10/$50 per 1M. And subsidized usage with Codex provides a greater budget on all tiers, and there are no restrictions on Deep Learning and AI/ML work.

[Vano @vano_ai]

Switching to https://t.co/M7udj1x06A as a harness, I validated that from the codex $20 plan weekly plan you get $100 dollars of subsidized usage. With this harness, the prompt is lasting seemingly far longer, but Sol (high) is definitely less efficient usage wise than with 5.5. I've noticed the main issue is reading outputs from commands drains the budget. It just reads the full outputs unchached, whereas Opus 5 tends to append a grep command to the outputs and has far more usage as a result.

[Vano @vano_ai]

@MizoChris Yeah, price will only keep going up. Personally I don’t need it to be portable, and at $1000+ there are better options

[Vano @vano_ai]

Progress on the Minecraft web port mod/harness.

[Vano @vano_ai]

@dukeofdolma Bias. I don't like Anthropic as a company but they fixed their usage and Opus is genuinely good. The browser/computer use isn't as good as codex, and Sol tends to work better for my use cases, but at least I can use the $20 claude plan for more than 3 hours.

[Vano @vano_ai]

I am unable to use >80% usage with Opus 5 medium in 5 hours, and if I do its only 5% of weekly. In comparison, I just ran codex for around 2 hours on Sol High and my weekly is at 40% from 100%. Both on the same codebase and Sol medium consumes about the same anyway. I will say after the latest reset efficiency is a little better but still bad.

[Vano @vano_ai]

@CodeAkram @YousufKalim Is it genuinely better on different harnesses?

[Vano @vano_ai]

Hmm, Opus 5 medium with a really large prompt (a trace that took 700k context tokens) filled 40% usage immediately, stopped only by running out of the 5 hour limit.

[Vano @vano_ai]

@rezoundous I wonder if by mistake they swapped the 5hr limit and the weekly limit when they removed it. Claude code $20 plan I can’t even use >80% of the 5 hour limit before it resets with Opus 5 medium.

[Vano @vano_ai]

@HardwareSteam Steam machine email/availability was at 11:09 AM PST for me when it came out, so I’m waiting another hour, else no hope for this thing coming out soon.

[Vano @vano_ai]

Opus 5 medium (most cost effective to intelligence level) can run for the full 5 hour limit only hitting 82% (in my experience) on the $20 plan before resetting. Best part is, if you use 100% of the 5 hour limit, thats around 5-10% of the weekly. So you have a lot of headroom. This used to be the case with ChatGPT 5.5 and not with Opus 4.8, kind of sad to see.

[Vano @vano_ai]

@Pirat_Nation Ig 26.3 or later will have some crazy optimizations. For a while now it is actually been detrimental to allocate more than 8gb of ram to it.

[Vano @vano_ai]

@IamMiaChase Burned through 95% weekly usage on the $20 plan in around 4 hours (sol medium, high for 1 hour maybe). Decided to test anthropic $20 plan. Opus 5 high, been going for nearly 4 hours now, 75% 5 hour limit used, and 8% weekly. Something is very wrong.

[Vano @vano_ai]

@uwukko If you mean fullscreen no, if you mean taking the full monitor space yes.

[Vano @vano_ai]

@argofowl The voice model for me has started to speak with a speech impediment (I don’t have one so this is random not mimicked) and audio glitches. Not to mention loosing the multilingual capabilities.