try-working

It's not China. It's individual labs. Open source is their go to market strategy: https://try.works/why-chinese-ai-labs-went-open-and-will-rem...

No, open models is a marketing strategy

This is their motivation https://try.works/why-chinese-ai-labs-went-open-and-will-rem...

sounds like a caching issue, and maybe other issues too

everyone is using OpenAI and Anthropic in China. We have both providers at work as well.

in case someone wants to speculate in why chinese labs open source their models: https://try.works/why-chinese-ai-labs-went-open-and-will-rem...

input lag is a minor part of it. it depends on the game, but for example in CS2, refresh rate changes game mechanics like recoil, movement, even trajectories of thrown grenades.

i went from 60hz to 240hz, with <100fps average, and the difference was still immense. refresh rate is more important than fps, that's how big it is.

it's recursion

no, you download the runtime and connect your models to it, then you use the router as the endpoint in your coding agents

you could use my model router to route between models like that. https://github.com/try-works/role-model

It's a GTM strategy: https://try.works/why-chinese-ai-labs-went-open-and-will-rem...

He was made to leave.

Absolutely not. OnePlus was more known and trusted outside China and should have been the flagship. OPPO has "the smell of factory girl" inside China and in SEA.

I worked in the OPPO and OnePlus HQ in Shenzhen. When I had contracts to sign, I wrote my own signature and then I had to go to some procurement person and get them to stamp it with a seal.

I worked at OPPO and OnePlus when OnePlus was being re-integrated into OPPO. It used to be its own org with a separate office, but the team was moved into the new OPPO office.

If you want to understand why OnePlus has been destroyed since then, there's a number of factos that I will just quickly list: OPPO was doing a large international push from 2019, and OnePlus was the little sub-brand that was more successful than it's mother brand. OnePlus was an international team, OPPO fully Chinese, with a CEO and the entire upper management not able to speak English. OnePlus continued to succeed while OPPO couldn't do good marketing. This was embarrassing, so of course OnePlus had to be downgraded. Pete Lau became OPPO CPO and decided that OnePlus should be a lower to mid brand, and the product was changed to reflect that. Lots of people left.

Ego, incompetent management that knew nothing about the world outside of China and terrible decisions that followed, led to the erosion of OnePlus.

It could have made them a real player internationally but they intentionally destroyed it.

open source is a GTM strategy

I downloaded two of the official ones in LM Studio, both 3.6gb, and neither loaded.

I feel that skills should live in the repo folder so they can be used by everyone on the project.

You should make sure to not read Stratechery then. It's writing is even worse.

they said it will do worse on that

the remaining 10% accounts for the other 10,000% of the development time

I'm guessing you mainly think of planning and maybe review when you say that Opus and GPT vary in their output. This is more of a qualitative judgment because it's difficult to write a planning benchmark. If you wanted planning to pass between both of these generalist models, role-model supports this via the pi-role-model extension for Pi that lets the coding agent send request metadata, define routing strategy to use and even model ID.

To implement it, you would make a rule for Pi to do another pass for any coder.planning task that it sends to the role-model router.

These types of personalizations are suited for implementation in the consumer application layer instead of in the router in my opinion.

Feel free to take a look at the docs for how routing decisions are made: https://role-model.dev/

When you use Pi with pi-role-model, Pi will include task and role metadata with its request; the role-model router runtime additionally holds benchmark and observability data, and a configured routing strategy. A composite of this is used to make the decision.

What you are pointing out is correct: making the actual decision and ensuring it is accurate can be difficult, which is why you need rich data as above, and also a model pool where each model is distinct, as I wrote in this post and in one of the comments below.

This is an interesting comment because it makes me want to ask the question, what is the use for fable then? For me, GPT 5.4 is enough when using recursive-mode. I do appreciate GPT 5.5 Pro for some larger research, architecture, planning tasks though. I think that's what Fable is for. A very small % of total work.

I can add to the above that an accurate model router is what enables specialist models, and specialist models is what will in turn make model routing common place.

When we have a standard model routing protocol in place used by both applications and providers, we can start to really reap immense benefits from accurate routing and fine-tuned specialist models resulting in better performance and lower cost.

a man of taste

I am European, working in China for 12 years making multiples of the average salary, speak Chinese above B1 level and am not eligible for permanent residency yet.

First of all, if you want to become a resident somewhere you must learn the language. Not should.

Second, no country owes any foreign citizen residency there.

well, I did go to 中央美术学院 so I have some sense of it

it's arty

If we define GPT 5.5 and Opus 4.8 as the frontier models for simplicity, there is some value in routing between them theoretically because two models will always have some differences.

However, when the models have the same generalist profile capabilities and are at the same performance and cost tier, making a decision for when to route between them and also making sure that that decision is correct, requires enormously granular information. While there are benchmarks that show differences between the models across different domains and tasks, the differences are generally not major and we also cannot assume that benchmarks that we know are optimized for, because if the new model wasn't presented together with good benchmarks the business would tank, really reflect real-world task performance at the request-level.

So routing between similar models is an information problem that is unlikely to be solved.

Routing between these two models is also likely to have a lower benefit than routing between GPT and DeepSeek on the cost vector. Routing to DS has clear, known and verifiable impact on cost. There is no need to guess.

Similarly, if we routed between GPT and a specialized math model, lets say Leanstral, that we can assume outperforms GPT by >50%, the benefits are also massively larger, and the routing decisions are also easy to make.

This is why the biggest pay offs come from routing between models that have a 2-10x difference in one of the cost-speed-quality factors, or specialized in a specific domain, or runs locally for data-security sensitive work.