AM1010101

It’s interesting they skipped caching. I could see wanting to ask follow up questions so having your first x tokens in cache would be interesting.

Also if you have a long “system prompt” then caching would have saved a considerable amount on bulk data processing.

There may well be a technical reason I don’t understand.

How well calibrated is it? Is 90% calibrated to be correct 9 times out of 10?

I think Jev had put significant effort here and its not clear if luna will be well calibrated in this way.

Half price on open router right now

Matches Astra on Artificial analysis at lower cost of $1.99 per task instead of $3.26. Still far more than GPT 6.1 sol at $0.79 for 1 point lower in intelligence.

I have found gemini models to have some of the nicest and easiest to read prose so I’m looking forward to trying this out. I hope the UI design has been preserved too

For me I would like to pair this with Opus 5.5 as orchestrater and use Sonnet as a sub agent. Therefore I want it to be fast when on low or medium and not break the bank.

On low and medium it seems competitive, maybe slightly cheaper than opus, in terms of intelligence per task.

If the time per task is lower (Artificial Analysis don’t have the date up at time of posting) then I have a clear use case for this model all other things being equal.

What frustrates me about many projects is it's never been easier to just make a quick video or image or gif and put it in the README to show what the value is. This was at least somewhat complex before AI but now you can just paste an image and have the AI put it in your README.

I guess the frustrating bit is that the readme is 100s of lines long. AI is already saving you so much time on a project like this but you still make me clone and spin it up if i want to see it for real.

I get that open source is a gift and you’re not obliged to do anything I say but please consider taking the extra few minutes on your next project.

Also just a tip - a picture, video or gif would go a lot further on your landing page than a wall of text. Show, don’t tell.

What are the envisaged use cases here? Miro to drawio?

Awesome, keen to see where this goes. Can we have light mode for the whiteboard and dark mode for the code?

Did 4.6 not have an x-high reasoning level? Why are they comparing 4.7 x-high with 4.6 high?

Would the idea be to have many specialist models (forms, wikipedia, final cut, etc) and have a parent model choose which is best (potentially also a specialist model for selecting specialist models) with the idea an llm would give a larger goal and trigger this cascade of specialists to quickly do the task?

This is a pretty good move for Meta to go all in on. AI is ChatGPT and Gemini to normies, Meta is not considered. However when I ask one of my less technical friends if they connected any tools (email, calendar, docs, etc) they are clueless that this is even possible. To them AI answers questions and generates images. If Meta can get this to the masses as the first AI thats useful at doing agentic things, ie an assistant rather than a chatbot, then the average user will see meta as as the company who provides this to them. It’s their best chance at winning the category and having a strong play in AI. If I were Chat or Gemini I’d get something to market fast. Timing wise are they also hoping to beat / be compared to any siri announcements this week in the media too?

I don't understand why google are being like this. This is an opportunity to extend an olive branch to developers (like openai has) and a great way to get data around how people are actually making this model work for them regarding coding…

Seems to do reasonably well in opencode according to artificial analysis. https://artificialanalysis.ai/agents/coding-agents

If I had to pay per token I would probably consider using this (they seem to be on the pareto of performance) but not being able to use opencode with a subscription is not really something I'm realistically going to do when claude and codex are around. Also never gotten along well with gemini-cli / antigravity-cli.

Interesting how quickly they seem to have caught up.

50% off at open router is also still applied so it comes out at $2 / $10 per 1M.

Feature request for Artificial Analysis, allow us to see these live prices on the pareto. It would amazing to also see what a 25,50,75,100 % utilised subscription costs compared to raw tokens.

Not true. I have used Fable for all my planning work in a reasonably large but well organsied codebase and I rarely hit a session limit. I would use all of the Fable specific usage if I accidently let it do the full implementation of the plan. Pro was perfect for me. I would then used GPT-5 / Composer 2.5 for implementation followed by a Fable check over. I will definetly see if the API usage in Cursor is enough to cover my Fable needs before upgrading.

Plannig and review. You want the best plan and when its done you want to catch all the bugs. Implementation can be almost anything.

Of course you can do the planning and reviewing by hand but sota models do a pretty great job, especially in review (please still read the code).

PLEASE MAKE UP YOUR MIND.

I have claude, codex, cursor and kiro, all most basic teir and swap around. I like anthropic the best. I would have upgraded to Claude Max 20x by now (and dropped the others) if you would just incldue an amount of fable in the subscription. I would have done it on the first of this month if I'd know fable would be included until the 19th. However I'm not sure I want to at all as I feel you will just rug pull me or string me along in another way and constantly be threatening to change things. I have other stuff to think about, the last thing I want is my tool provider constantly changing their mind on things, making me waste mental energy on -p or fable access.

I don't really want to but it would be so much easier to just go with codex right now...

I would love a service that I could upload these chats to (anonymously) so that those developing open models can have it as training data and not just the closed model companies. My understanding is that it’s very valuable, look what cursor have managed to train. Obviously some filtering so that only chats or projects you want to share get shared would need to be in order.

Is that not worth running the benchmark on to prove or disprove this anyway? This would send a strong signal to google to get their act together and save me from wasting tokens selecting high.

Out of curiosity how are benchmark runs generally funded? It would obviously be great to test them all on all reasoning levels and in and out of their native harnesses. Maybe even in pi / opencode / cursor but I get this would get prohibitively expensive unless you have funding or free tokens.

Thanks for your efforts thus far. Looking forward to seeing more.

Im just replying to my own comment to praise the team for the latest update. They have fixed the speed issue and the text is now a sane width on my screen.

Well done and thank you.

If I was to ask for more there are a few things that Claude and Codex are doing better on. Way more integrations / skills. I wish I could use MCPs and skills here. Modern design and feel too. This design, while native, feels very outdated

I was essentially installing the web app as a PWA. Why? This allows me to set a deterministic keyboard shortcut to open gemini, in my case super + C, as if it were a native app. Downside is that it eats some ram. I haven’t benchmarked this I have no idea how heavy on ram this really is. FWIW I’m fine spending ram on this if it improves my workflow, which it does.

I've had the gemini web app in a safari embedded browser tab for 6+months its worked great as a local app. (M1 Pro base spec) I set Super + C set to open it.

First impressions of the native app: I think I will save RAM and it will be faster - Cool. Download, install, sign in - Easy. Change the keyboard shortcuts - Simple. cmd + N to open a new chat - better than cmd + shift + O in the web app.

Back to work, a few moments pass.

super + C - ... wait ... there is a noticable delay in opening the app?

Maybe try the new mini chat? - Same delay 800+ms?

Thats annoying, every other app on my mac switches instantly?

Why is this slow? The delay long enough that you break my flow and make me think spending the RAM is worth it for the old web version? Whats the point of native if its slow?

Second grip - the text layout is too wide to read comfortably. At least give me the option to put it in a more narrow layout. I use every app full screen so reading text across the full width of my screen is pretty awkward.

Anyway I'll give it a go and see if it grows on me but right now the web app feels more polished and responsive so I will likely switch back.

Edit: typos

Awesome work, keen to try it out tomorrow. Can I make the notifications work with Gemini CLI and Kiro CLI too?

It would be great to be able to kick of a chatbot convo to go deeper on something each time you have answered. Chat for a bit to deepen understanding and read around, then pop back out and continue to the next question when your satisfied you've understood the concept.

Agree on colours. The Electronics topic was white on yellow, completely unreadable.

Nice chrome apps. What do you use to build them? How do you find the process?

I live in one of the colourful houses in Clifton wood. We have a local WhatsApp group and one of the residents posted this a few months back: “A couple of people claim they were the first to paint Cliftonwood houses a bright colour in the 70’s, one being our ex-mayor George Ferguson. But the wonderful late Joan from Cliftonwood Cres told me that was ‘codswallop’. Also when we moved in here fifteen years ago, we were told there were five colours allowed for the houses and that was that .” Others then disputed it and said there was never a 5 colour rule and quoted Annie Scott (who was an architect) started the trend.

I’ve also heard it said that this area was popular popular with artists and the rebellious types back in the day vs the more posh types in Clifton on top of the hill. There are some photos of the harbour from back when it was a working harbour, before the houses were painted, and it looks far less appealing (https://www.instagram.com/p/CyMNzAXIOY-/?igshid=MzRlODBiNWFl...). If you do end up on Instagram search for Bristol harbour to see what it’s like now.

Most of my life I was like this. I studied for exams the night before, same for coursework. I did the night owl thing. I would leave my work right until internal deadlines and rush or work late to finish.

9 months ago I started a new job. It was a relatively senior role for me. Straight away my manager kicked my ass if I didn’t get stuff done. It was fully expected that I was a productive and competent member of the team. The prospect of loosing my job was very real. There is also no working from home so there is nowhere to hide. I quickly started focusing and putting in extra effort. For a while I had to put in extra time outside work to get up to standard but now I just work productively and go home.

When I work on my side projects I make far more progress far more quickly. I feel far better about work and life in general as I don’t feel like I’m wasting my time.

My advice is to put yourself in a situation that will demand a lot from you. Make sure you can’t make excuses and that you have someone/thing to hold you legitimately accountable to a high standard. It may be hard but in the long term it’ll help you reach your potential.