AI Judge changed title after evaluation, original title: "Factory launches Factory Router, an automatic model routing system designed to cut AI inference token costs by 20% to 25%"
The system dynamically routes tasks to smaller, cheaper models.
Not all tasks need a frontier model
People have been talking about dynamic model routing for years
But it's finally here
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Many users praised FactoryAI's model router for automatically selecting optimal models to cut enterprise AI costs 25% while preserving performance, calling the token optimization a game changer.
No Digg Deeper questions have been answered for this story yet.
Every enterprise CFO is going to demand this.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Model routing is an important thing
Controversial idea: the frontier labs will want their AI harness to be the moat, but ultimately the best case for consumers is that model capabilities flatten and commodify
Preview of the AI Harness Wars of 2027
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
👏👏👏 @matanSF
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
💯
Each enterprise I've spoken to is spending their AI token budget quicker than they anticipate... @FactoryAI and @matanSF just came out with model routing which will help lower costs and save tokens without sacrificing performance. Will be huge as AI spend continues to explode!
As token budgets take on a larger part of operating expenses over time, model routing is the inevitable conclusion. This is also one of the biggest areas of differentiation for the applied AI layer over time.
By understanding the different work patterns in your domain, and having strong evals for that domain, you’ll be able to cost/performance optimize effectively.
We’re still likely at the point where most use-cases will need frontier performance for the foreseeable future; but soon you will be able to peel off individual use-cases and send them to lower cost models once the quality is sufficient for the task.
Enterprises individually trying to figure this out themselves at scale will likely not be possible, so the products that can intelligently route these workflows to the right tier of model will be in a strong position to aggregate more demand.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
If your daughter needs tutoring in algebra, you can probably find someone cheaper than Albert Einstein.
Giving every task to GPT5.5 or Opus 4.8 is overkill. Often times you can get the task done just as well, but 10x cheaper and 10x faster with a smaller model.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
This is one of the most subtly important releases we've made.
Models are ultimately commodities. Having a liquid market for them let's us achieve performance that is strictly better than any frontier model on every dimension that matters: price, quality, and reliability.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
This is actually one of the main advantages startups have over frontier labs, as long as there's a healthy spectrum of open-weight to closed-weight models on the cost-performance curve.
Building a model harness is fundamentally a bit different than building the underlying model. The last year has shown that an increase in model capabilities has led to an increase in sophistication around agent harnesses as well (e.g. just look at Claude Code ultracode)
Every startup has a meaningful advantage here - from Factory to Cognition to honestly us - if they're able to optimize and finetune the model set for the precise amount of intelligence needed to do a task. Same or higher accuracy at 2-10x lower cost and latency is a real advantage.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Automatic behind the scene routing in user interfaces (instead of model picker) will redistribute value capture and usage towards many more models than just frontier ones (especially towards open-source/smaller/cheaper ones). Because it removes the cognitive load for the final user to have to switch models (which is too high leading to defaulting to frontier for all calls with model picker) cc @antonosika @bgurley
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
this is pretty genius. in a world of increasingly expensive and abundant ai models products like this are a dream
AI model usage has shifted in a huge way recently:
- flat fee subscriptions to usage-based models
- single model usage to multi-model prompt routing
why? subscriptions were too heavily subsidised and different models are good for different things
every fortune500 company has BOTH a claude AND chatgpt sub. engineers are using claude to write a plan and codex to execute
factory router claims to save you 25% cost while maintaining frontier results.
if you’re a company spending $100M per year on tokens (and there will be many) then saving $25M is a a no-brainer
for routing companies like factory they then build a huge moat around intents.
v smart (and no, i’m not an investor, i just like the tech)
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Routing the right model to the right task is exactly the kind of unglamorous, high-leverage work that compounds. Factory is helping make this the default, not a tradeoff. Great to see @matanSF and the Factory team help builders optimize for both quality and cost.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
What needed a frontier model last year runs on a more efficient one today, yet AI costs keep climbing. A higher token bill doesn't mean more work is getting done. Engineers default to the most powerful model out of fear of losing performance, so routine work runs the same premium path as the work that genuinely needs the expensive model.
Factory Router cuts token spend by 20-25% while maintaining frontier performance. It automatically selects the right model for each task, and routes across providers for reliability. Designed for agents, it switches models only when the gain is worth rebuilding the prompt cache.
We mapped the cost/performance Pareto frontier across benchmarks. Near the top it's nearly flat: cost drops sharply while performance barely moves. Then it bends hard: the most aggressive routing we measured cut Terminal-Bench 2 to 56% of Opus cost but dropped pass rate to 81%. Factory Router operates on the flat stretch, right before the bend.
As efficient models improve, more work crosses into what they handle just as well. With automatic routing, users will see growing savings at frontier performance.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Model routing is not that easy for adhoc chats. These prompts are too short and far from what we see in the real world. Only when you have a full prompt for a wyorkflow, model routing makes sense. Getting prompt caching is very important as most caching saves you 60-80% cost. You can't simple route each message in a chat to a different provider let alone different model. Each model has very different restrictions on images, tool calls, etc so you need to choose the model family and stick with it.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
Potentially dumb question, but doesn’t cache busting mean this is actually worse for anything that you can’t one shot?
1st prompt Yay cheap model 2nd prompt Oh wait we need the big model, which means we resubmit your entire chat history
So you pay full price for 2nd prompt, which includes paying for the uncach’ed preceding history, which means you now paid twice for the first prompt, once for cheap model and once for big model
The benchmarks are very misleading here because they only test a single prompt sent. But if you need any sort of follow up you are paying extra
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
This is absolutely true.
Large foundation models save on labor costs at the expense of operational costs.
It's an "Exploration-Exploitation" problem. Explore with large, general-purpose models and Exploit with more cost effective ones.
Automatic is the best of both worlds!
Every enterprise CFO is going to demand this.
Best quality, lowest cost. Pick two.
Factory Router is in private research preview today, in the Factory CLI and Desktop App.
Read more here: https://factory.ai/news/factory-router
Thoughts: 1. Tech twitter/SV severely underestimates the accelerating financial pressure that the average person/business feels around AI spend 2. Gap between frontier and OSS models decreasing rapidly 3. The US needs OSS supremacy
Aura is clearly shifting to @FactoryAI
Each enterprise I've spoken to is spending their AI token budget quicker than they anticipate... @FactoryAI and @matanSF just came out with model routing which will help lower costs and save tokens without sacrificing performance. Will be huge as AI spend continues to explode!
Factory just introduced Factory Router, a coding-agent model selector.
Claude Opus-class results while cutting AI session spend by 20-25%.
Reports 99% of Claude Opus 4.7’s Terminal-Bench 2.
Basically it works by treating each coding-agent run as a routing decision: it first sends the task to the cheapest model class that should be strong enough for that kind of work, then escalates to a stronger frontier model if the session starts failing or needs deeper reasoning.
Frontier AI should be reserved for frontier work.
Introducing model routing to Factory.
Factory Router picks the right model for every task, automatically.
Maintain frontier performance while cutting costs by 25%.
A higher token bill does not mean more work is getting done.
One-line fix? Doc update? Too often, routine tasks get routed to the priciest path out of fear of losing performance. This only burns budget for no additional gain.
You wouldn’t have Messi play goalie.
Every model has different strengths whether it's reasoning, speed, cost, or context. Factory Router automatically picks the right model for every task.
Factory Router delivers frontier performance at lower cost.
Benchmarks: • Terminal-Bench 2: 99% of Opus 4.7's pass rate at 20% lower cost • Legacy-Bench: 96% of Opus 4.7's pass rate at 25% lower cost
