Technique of using a frontier model for planning and a cheaper good-enough model for code execution. The creator's worked example cuts cost from $9.50 to $3.02 (about 68%).
Model routing
Technique of sending each AI task to a different model, e.g. a frontier model for planning and a cheaper one for execution.
Updated EU status checked Last shown
Model routing at a glance
- What is Model routing?
- Model routing is a research. Technique of sending each AI task to a different model, e.g.
- Does Model routing work in the EU?
- Unverified. No explicit evidence found. This is a general technique rather than a regional product, so availability depends on the models and tools used. Last verified 14 Nov 2026.
- How much does Model routing cost?
- Pricing unknown. Not a product; the technique itself has no price. Costs depend on the models and tools used.
- Where was Model routing shown?
- In 1 creator video by Matthew Berman. Most recently by Matthew Berman on 7 Jul 2026, at 0:00 in “Cut your AI cost IN HALF (EASY)”.
Sources: creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Model routing is a cost and quality optimisation technique, not a single product. Requests or sub-tasks are directed to different language models depending on difficulty, type or stage of the workflow. Simple or well-defined work goes to a cheaper, faster model, and harder work goes to a more capable frontier model.
A common pattern, described by YouTube creator Matthew Berman, is to use a frontier model for planning and a cheaper "good-enough" model for code execution. In his worked example, the cost of a task fell from $9.50 to $3.02, a reduction of about 68%. This figure is a single creator-reported example and will vary by workload, models and pricing.
Routing can be implemented by hand (choosing models per step in an agent or coding tool), with rule-based logic, or with a router or gateway that picks a model automatically per request. Trade-offs include added complexity, the risk of a weaker model failing on tasks misjudged as easy, and the need to evaluate quality across the models used.
- Assigns tasks to different models by difficulty or workflow stage
- Planner/executor split: frontier model plans, cheaper model executes
- Reported example cost cut from $9.50 to $3.02 (about 68%)
- Can be manual, rule-based or automated via a router or gateway
- Needs quality evaluation to avoid regressions on misrouted tasks
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Similar AI tools
Closest neighbours by meaning, not just by category.
Not Diamond
API · Not Diamond
Intelligent model router API that picks the model and reasoning effort per step for coding agents to cut inference cost.
GPT-5.4
Model · OpenAI
OpenAI frontier language model used for coding, reasoning and agentic tasks, available via ChatGPT and the API.
GLM 5.2
Model · Zhipu AI
Zhipu AI's GLM 5.2 large language model family, with Fast and max variants, available via Z.ai chat and API.
Muse Ember
Model · Meta
Meta's smallest Muse-series language model, positioned as a low-cost option for subtasks and delegation.
Runway Media Router
API · Runway
Runway API routing layer that picks a video, image or audio model based on cost, quality and latency targets.
GPT-5.6 Luna
Model · OpenAI
OpenAI's lighter, lower-cost GPT-5.6 model variant, suited to routine tasks such as setup and scripting work.