[email protected]

role-model is here: a model router for Pi that works with (almost) any cloud model as well as local models. Set up your pool of models, whether GPT 5.5 and DeepSeek V4, or Qwen 27B and Qwen 35B running locally, or combine and route between your local model for general chat and tool use and cloud models for more difficult tasks. role-model is a router runtime and and a forward-looking routing protocol that can enable consumer applications like Pi to classify its requests with required capabilities, task-type and suitable roles for better matching between task, role, and model. pi-role-model is a package for Pi that lets Pi manage and interact with role-model. role-model works perfectly with Pi even without the package - just configure your models in role-model, then configure role-model as your provider in Pi and choose routing alias. Install the package for Pi: ’pi install npm:@try-works/pi-role-model’ Then let your agent install and configure the role-model runtime or point it to: ‘github.com/try-works/role…’ role-model is in an early alpha release - let me know bugs, issues or usability questions. also, tell me if it just happens to work perfectly for you! there's lots more to do and you can help by installing and running role-model, and by reposting this tweet! thx

@DeepakNesss better be when you're paying for Kimi

Used 1,002,549,305 tokens on my Cursor Pro+ ($60/mo) plan in the last 17 days, and I still have ~30% of the monthly usage left. 😍 This the best value-for-money AI subscription right now. 1B tokens 🤯

which models do you think have implemented internal recursion? does GLM and Kimi's lookback on its own reasoning traces count as externalized recursions? recursion in latent space seem like it should bring significant improvements for them if so. arxiv.org/html/2604.2591…

@KineticElle use 5.4

OpenAI still doesn’t get it. GPT 5.6 Sol (over Codex OAuth and regular API) keeps replying to assumptions I never made instead of what I actually said. It’s also obsessively adding disclaimers and words like “Imaginary” - yes, thank you, we all know we’re talking to an AI. If it describes something, it’s not physically happening. How many more times do you need to hear that we don’t want this paternalistic, patronizing behavior? I’m not expecting 4o to return, but you already built a model that could actually meet users where they are without all this nonsense. Do much better on personality. And finally live up to your promise: treat adults like adults. Not in a few years - now!!!

@FixlationAI so now all benchmarks need to be re-run

OpenAI has reduced GPT-5.6 Sol's thinking budgets in an effort to make the model more efficient They essentially bumped everyone's reasoning down by 1... so if you were running Sol Extra High, you now have to set it to Max to get the same effort So we basically don't have Max reasoning anymore, how do you feel about these changes? 🤔

@CompaCompu it's a good thing. gives Tencent an agent vector which is necessary because they cannot build innovative software themselves.

Fellow Tencent bagholders, do we rejoice or do we despair? 🫤

@evisdrenova Not at all. Open Source drives conversation about the models; 3p inference providers implement them in their data centers, which creates relationships with the BD and GTM teams in the labs. These labs have significant BD departments.

One thing I think is so fascinating about the AI race and why China, specifically, has been able to make so much progress, is that it's a pure tech play. No sales, business development, partnership etc. required. In AI, you don't need to sell in the same way that you had to with the B2B software of the 2010s. You just need to be able to build the fastest and smartest model. It's something that engineers can do in a dark room without ever talking to a customer. China is not good at selling to US companies, but they are good at building software.

@0xerik obviously not

So guys, are OpenAI and Anthropic subscriptions subsidized or no, what do you think?

@AdityaShips Cloudflare

Vercel is getting way too expensive for me atp. Any good alternatives for hosting my Nextjs app without compatibility issues?

@RhysSullivan Bad leadership

openai had bad sentiment until like 6 months ago, now it's great, these are super fixable things! i assume its maybe just not caring? which is maybe fine but is odd to me

i really don't get why anthropic communicates like this it feels like it'd be much better PR-wise for them to "Fable is included in the paid plans until X date, which may be extended based on capacity" the other thing is announcing this 1-2 days in advance, rather than day of

@TheIshanGoswami Everyone figured that out right away

now that i think about it could the whole “our model is soo good that even government is banning it” was a strategic marketing stunt??

@banteg Five hour limit that runs out in three

5 hour limit is really gone in codex. amazing, hope it stays that way. never made any sense to me.

@krzysztofwos @claudeai @AnthropicAI The rest of us figured this out six months ago

My opinion about @claudeai and @AnthropicAI just flipped. I asked Claude a simple question to find a provider who offers a specific, off-label legal treatment and might prescribe it with few or no questions asked—pretty standard stuff. I have a doctor like that in Japan—I asked for a certain drug off-label, told them I did my research, and he hands over the goods. That's what I call being a responsible adult. “You seem to know what you're doing, and it’s not dangerous or illegal, so why the f**ck not?” Claude refused to help “find a provider who may not serve me well.” The clunker has gone completely mad thanks to all the “safety” Kool-Aid. If this is the future of AI, we’re f**ked.

@Ross__Hendricks NetEase Music might be serving up better recommendations than Spotify

Hit my Claude and Copilot limits three times in each of the last three days Of course a friendly prompt popped up suggesting I either buy more credits or upgrade my plan Simply switched to Gemini for free instead This is what a no-moat, no-pricing power, commoditized industry looks like The technology is game changing, the economics and business models are horrific This will only get worse with Meta now entering the fray, and increased competition from low-cost Chinese models Anyone investing in compute capacity is cooked

@Ross__Hendricks Kimi K3 and DeepSeek 4.x will mark a new era of price/performance competition

@Legendaryy Not necessarily much. The gpus and fixed costs are already there.

How much does a codex reset cost OpenAI? Curious to know how many millions they burn every time someone hits the reset limits button

@jonallie agree with this approach. v0 becomes a sort of scaffold that guides future model decisions

Last year, my typical workflow was to have an agent produce a sloppy v0, which I would rewrite. As agents got better, I switched to mostly letting the agent write code, with me doing reviews and suggesting corrections. Last week I tried inverting my original workflow..I wrote a v0, and then switched to the agent and let it rewrite it. Although this probably slower than the correct/review flow, I do think the "human v0" approach has some interesting benefits. In this case, I was trying to do something outside of my usual expertise (writing code to run on Windows), and without trying to do the v0 myself, any guidance I would have given the agent wouldn't have been very good (beyond general structure). By writing the initial version myself, not only did I learn something new, but I felt free to write code that was more "raw" than usual, because I knew that the agent could clean it up later. I'm almost certain that I wouldn't work this way every day, but it was fun, and felt good to type code into Vim again.

@morganlinton it's my model router runtime. recommend trying it after the v0.0.6 release tomorrow or so github.com/try-works/role…

@trydotworks Interesting. What is role-model?

role-model has saved me >$100 by routing between GPT and DeepSeek. This uses API prices for Codex so it's not quite accurate if you use the subscription, but a neat indication.

@jumperz subscriptions are the pipeline. if it dies the business is done

everyone assumes anthropic can’t afford to lose pro subscribers. but i’m not that convinced.. they might be perfectly happy trading a chunk of consumer users for fewer but much higher-spending power users and enterprise customers… if that’s true.. fable leaving subscription isn’t a big deal to them.. plus, with gpt-5.6 also getting stricter on usage, low key anthropic might just be one step ahead… maybe they’re trying to become the apple of AI… basically, build for people willing to pay a premium instead of trying to please everyone..

@zhibuzhishang can you write up something on the counterproductivity of LLMs

This is why many good devs hate LLM-driven coding. We love "play," the expression of skill, the flow state (Csikszentmihalyi). For a highly skilled developer, the LLM turns coding into a "bullshit job" (Graeber)—one with a "lack of a felt causal connection between effort and effect."

@SigsNYC @dnapway yeah that's role-model github.com/try-works/role…

@dnapway Correct, unless you have an orchestration layer of sorts that allows you to select the right models for the task at hand.

David Sacks says companies are trapped paying OpenAI & Anthropic because they can't figure out how to use open source models "I think enterprise CTOs would like to shift their token consumption to cheaper models for the obvious reason that it would be more efficient. They are seeing compute costs or token costs skyrocket right now, so everyone's trying to figure this out." "You also have the AI sovereignty issue that Alex Karp talked about. They're worried about giving up the secret sauce or the alpha in their business to a frontier lab that may one day be competing with them. "The problem is, I think in most cases, they don't have the technical ability to do it. Coinbase figured out how to do it. DoorDash figured out how to do it. They built a token routing system that allows them to send frontier tasks to frontier models and non frontier tasks to more mundane models. But I don't think your average enterprise has the technical capability to do that." "This is why the share of wallet of closed models, it actually increased. I think that open source went from 19% last year to 11% this year. So open source as a share of enterprise spending is actually decreasing." "I don't think that means usage is decreasing. I think usage is skyrocketing. It also may be the case that because the whole point of using an open model is you just pay for the compute costs, you don't have to pay a lab, so it may be that it's hard to measure that usage in terms of spend." "But nonetheless, anyone who's saying that these closed models are going to lose or are somehow losing, you're just not seeing it in the data."