Comment by paxys
4 days ago
Looking at all these releases it’s not a surprise that model routing is the fastest growing segment in AI right now.
There are 10+ LLM companies, each with dozens of models of different modalities, each model with multiple size variants, then different “thinking” levels, then agentic modes, “pro” modes, a “fast” option, standard vs flex vs batch execution. And of course each end combination has a different input/output/cache token price.
Companies that say “give me a prompt and I’ll route it to the most ideal and cost effective model and setting for you” are capturing a ton of value from a gap that model developers don’t seem to understand exists.
Model Routing will always be done better by models themselves. Plus routing loses context making it more expensive and less reliable.
Model Routing is just Bitter lesson. The models themselves will get better at this and frontier companies will simply give that capability
This doesn’t seem obviously true, eg an Anthropic model will never route to Kimi even if it were best suited for a particular task.
I think what the parent is saying is that the model itself has the best context for whether a portion of a request should be routed. The specifics of that routing (e.g., should you route to KimiK2) are something that can be trained, finetuned, or even included in a model's startup context.
4 replies →
The only metric that matters is f(cost, time, task). If Anthropic has a SOTA model, it can easily distill every model in the pareto frontier and have the SOTA model route appropriately. They haven't felt the 'sting' yet to optimize that (still growing their ARR at a crazy growth rate)
Ant can top any benchmark that measures f(cost, time, task). The only entities that can beat them in costs are infrastructure providers who can do optimization at that layer. But a pure Model company can *never* compete with Anthropic on f(cost, time, task) if they continue to have SOTA models
Has anyone tried that? I have a feeling that if I put it a prompt Claude would comply. But I am all in on the Claude cool aid.
2 replies →
Why should it? An Anthropic model is architecturally optimized for Anthropic models, routing it to Kimi makes zero sense
3 replies →
[dead]
i think you're thinking of subagent routing
model routing in this case is cross-provider
Imo the main issue behind model routing is you need to figure out how much intelligence a new task takes, which is a very non trivial problem. Presumably, a organization knows this about their own tasks and is better suited to built in-house compared to outsourcing to a vendor.
> Model Routing will always be done better by models themselves.
> The models themselves will get better at this and frontier companies will simply give that capability
I would never trust something like model routing to the same company that would profit from it, and that goes for telling models doing their own routing when that could easily be trained into the model to make things more expensive. Sort of a conflict of interest.
Models are first and foremost trained by corporations.
Any different company providing the routing can be financially or otherwise incentivized to route this way or another.
2 replies →
Model routing by the model itself requires the model to pull in a lot of context and it's likely more efficiently just done by people with the context already in their head, even assuming the model is perfect at routing (which last I checked, Claude definitely isn't). I wouldn't trust ML model routers.
This is actually simpler to implement. You can have a AGENTS.md describing who is boot at what and then you have the agents converse over tmux.
Because they're trying very hard not to understand it.
Otherwise the expensive-yet-powerful model probably won't see much revenue. How much money is there in bleeding edge scientific research? There's a lot, but there's even more existing capital in paying people people to do college level paperwork, and the bulk of those traffic gets routed to the cheapest model.
You mostly don't need super powerful AGI to replace the paper pushers, but the frontier labs are trying to position themselves as being uniquely capable of producing super powerful AGI, and also be the ones replacing office workers.
Not sure how it will work out for them, but I think model routing is going to poke holes in that narrative. That's why I think they're trying very hard not to understand model routing exists.
It wasn’t long ago at all that the chief problem was “can AI even help me with this” (cost be damned). Until a time when the answer is an unmitigated “yes, obviously”, the frontier labs have everything to lose and nothing to gain on routing, because if they screw it up you might incorrectly decide “nope, it can’t help yet” due to a poor routing decision.
Is there really that much money in bleeding edge scientific research though?
This is what terrifies me about this whole ordeal economically. Maybe we get AGI and it is not worth anything close to what we thought it was for those who have a bet on it.
I think of what was the direct, economic value in the betting sense of quantum mechanics or relativity? Huge value at the systems level of society but as you scale down towards the individual the value is more and more dispersed to the point I would think any pool of bets would have all not paid off.
You can't monopolize and commoditize relativity.
I almost think there is a kind of dutch book against the AI equity holder because even in the best case scenario the bet doesn't pay off anything close to what is expected for an individual bet.
Who are the customers though? Honest question, I'd like to understand it.
For me, anything other than current best available SOTA for any task is unacceptable. The only routing rule I need is "the most powerful model I still have flat-priced quota available for". I mean, why settle for less?
There are two types of users: those who are able to use subsidized rates, and those who need to use API rates due to audit requirements, enterprise billing, etc.
Model routing for subsidized users takes the form of a "use Opus 5 subagents for implementation" type of system prompt. You lean into a single provider, build tooling around that, and your savings are far beyond anything multi-provider routing can get you.
Model routing for enterprises is far more complex - approaches like https://fireworks.ai/blog/kimik3-fable become necessary for cost control.
on top of that there is also additional factor: speed - sometimes if task is easy you do care to finish it faster.
There is also matter about convenience - when I ask some small easy question often I don't bother to switch the model or forget in prompt to ask faster/cheaper subagent.
It's very common to use a lesser model for a lesser task, resulting in same quality output. End result: save money while being faster. In many cases, it's a pure win-win.
But in other many cases you have to redo the work directly or indirectly, and you are more expensive (for now) than even the most expensive models, so sounds like a total lose.
I’m assuming you don’t pay per API call. Every mid-large sized business in the world does.
> For me, anything other than current best available SOTA for any task is unacceptable.
Then you must route. An article with lots of upvotes yesterday or two days ago showed that K3+Fable 5 was more SOTA than either of those.
Sometimes you can get an equivalent result in a fraction of the time using a less capable model.
The problem is of course knowing ahead of time that the faster model can give you the same result.
I'm not a customer of those routing systems, but I quite often use different Claude models for different tasks. While most tasks were Opus 4.8, I often used 4.8 to make a plan, prompts, and package kit to setup Fable for a bigger project, then run it on Fable. Or, for broad single-task searches Sonnet with or without "Research []" turned on seemed to work best both faster, lower overhead, and less verbose answers (when I didn't want it).
OFC, YMMV
And you do not have flat priced quota for Fable 5, right? Because nobody has, as far as I know. So you'll probably not route any task to the "current best available SOTA".
Also: quota. Implies you do not have unlimited access even for flat prices. Which in turn implies that as soon as you hit the quota on the most expensive flat price plan, even you will suddenly discover the magic of economically sensible behavior.
Fable 5 is included in the Claude Max subscription. I've gotten close to the limits this week and last but haven't hit it yet.
1 reply →
Not everyone is you. Other people probably have a range of tasks that can accomplished with different models.
Certainly if I'm confident that I'm going to get what I need from a faster model, that's what I want to use, rather than wasting time grinding away for the sake of saying of the same answer came from a SOTA model.
Given that every chatbot does offer a range of models, it seems clear people do choose among options.
The mental effort in estimating what model would be better is so not worth it.
I just want to switch to Claude Code, tell it to turn a .csv into a BigQuery table then cmd+tab to something else while it runs. Thinking "oh this is probably an easy task, I can /model to Sonnet to save $0.0004" is silly.
3 replies →
because "less" can be so much faster?
not just that -- "less" ($$$) can also result in indistinguishable quality for some tasks/inputs. I'd argue this is the primary reason, secondary being speed.
I generally agree. Perhaps there's only a 5% chance that it would write better code or find a bug that it wouldn't have with a lesser model, but the economics of bugs is strong enough that preventing a single bug is worth hundreds of dollars.
That's nice and all, as long as flat-priced quota continues to exist. Seems unlikely to go on forever!
lol I had to get ChatGpt to explain to me the difference between 5.6 sol, 5.6 Terra, 5.6 Luna, 5.5, 5.4 mini, 5.3 spark, and then there is low, medium, high, extra high, max, ultra, and pro… I still don’t really know, it feels like ordering hot wings.
And it's not just that model routing is much cheaper: no longer than yesterday we got a post showing that routing between K3 and Fable 5 was more SOTA than either of those.
If that is true, model routing is here to stay.
It also seems to validate the minimalist approach of pi.dev, where sub-agents from the same company is not the preferred approach (pi.dev believes in neither sub-agents all from the same company nor MCP even you can do it if you want for pi.dev's philosophy is to do add any functionality you want to a minimal harness).
Now of course we'll get for a few weeks all the Anthropic fanbois and shills explaining that "sure, K3 was basically at the level of Fable 5 but now that Opus 5 is out, open-weights models are six months behind".
How does model routing work if the prompt is static? Do the routers attempt to tweak the prompt to make it the best version for that model? For example, reading the “migration” guides for Claude Opus 4.6 -> Opus 5 is exhausting. Adding in all of the other models complicates it even more. Especially across vendors.
Openrouter should ideally kill in this space and make their model agnostic infra like memory, harnesses, chat applications.
OpenRouter is in acquisition talks with Stripe, fyi
I would expect routers to commodify like tokens.
OpenRouter sprang up overnight. I might need to replace some urls and access tokens should they decide to try and screw me.
1 reply →
what's the threshold for model routing where you're willing to trust the router?
For coding my own work I don't trust the model router, and it would have to be shown to be to save a real dollar amount.
From a buying perspective it's a hard sell to save x but lose out on bugs you are probably introducing at an unquantifiable severity and frequency. How much is it worth to hedge your bets by doing every single inference request on the frontier model?
How much will it cost to go back later and fix things, but also the meta question of how to be able to decide on a hypothetical unknowable? (You'll never know how much better or worse your code was gonna be, it's untestable at a project level)
you trust the service provider but not the router?
weird, but ok
I don't want to save money so badly that I'd possibly undercut the quality of the code that gets created.
*edit to add: that code quality (or lack of quality) is it's own cost
I am not convinced this will end up being a domain of the ‘routers’ vs the clients, as in harnesses themselves. Thoughts?