# Claude Code and Margin Pass Through: Pricing Lessons for AI Products > How Claude Code and similar AI products handle pass through pricing of model costs. When pass through works, when it doesn't, and what to charge instead. - **Author**: Ayush Agarwal - **Published**: 2026-05-19 - **Category**: AI, Pricing, Margin - **URL**: https://dodopayments.com/blogs/claude-code-margin-pass-through --- Claude Code arrived as Anthropic's own developer focused agent and immediately raised a question every AI product builder has been wrestling with quietly. How much margin should you actually charge on top of the underlying model cost, and when does pass through pricing make sense versus a healthier markup. The question has implications well beyond Claude Code. Any AI product built on top of frontier model APIs faces the same trade off. This article walks through what pass through pricing actually means in practice, where it works, where it breaks, and how to think about the right margin for your AI product. The framing is for SaaS, AI, and developer tooling, not ecommerce. ## What pass through pricing actually is Pass through pricing means charging the customer at or near the underlying model rate, with little or no margin on top. The product makes its money from the subscription anchor or from value added features rather than from the model cost itself. A pure pass through model would be the customer pays the exact provider rate plus a small operational fee. A near pass through model adds a modest margin, usually under fifty percent, that covers the seller's operational cost without accumulating significant per token profit. The contrast is with a healthy margin model where the customer pays two to three times the underlying rate. The seller takes the markup as gross margin, which funds the rest of the business including infrastructure, support, sales, and ongoing engineering. Both shapes work for some products. They have very different economics and very different sustainability profiles, and choosing between them is one of the more consequential pricing decisions an AI product makes. ## Where pass through works Pass through pricing earns its keep in a few specific situations. ### The product is a thin wrapper around the model Some products genuinely add limited engineering value beyond connecting the customer to the model. The user interface helps. The integration into the developer workflow helps. The model itself is doing most of the work. Charging significant margin on top in this case feels exploitative because the customer can see they are paying for a service that is largely provider passthrough. For products in this position, pass through pricing is the honest model. The fixed subscription covers the seller's actual fixed costs. The variable charge tracks the model bill. Customers feel the price is fair because they can verify it against the public rate card. ### The customer base is sophisticated and price sensitive Developer tools, AI infrastructure, and similar categories have customers who can read provider rate cards. If you charge two times the model rate, your sophisticated customers will notice and resent it. If you charge close to the rate, they will see it as fair and stay with you. Pass through is the model that survives contact with sophisticated buyers who do their own math. It is also the model that loses customers if you try to walk it back later. ### The model is provided by a sister or parent company When the seller and the model provider are the same entity, pass through is structurally aligned. The provider absorbs the model cost as cost of goods sold for the company, and the customer pays a fair price for the wrapping product. Anthropic's Claude Code is in this position. The model cost is internal. The pricing for the customer wrapper does not need significant external margin. ### The product is in a strategic land grab phase Some products price aggressively to win adoption with the assumption that pricing power increases later through stickiness, lock in, or category leadership. Pass through is one shape of aggressive pricing. The bet is that the customer relationships are worth more than near term margin. This works when the bet pays off. It does not work as a long term sustainable model unless the surrounding business has another revenue source that funds the operation. ## Where pass through breaks A few situations make pass through pricing structurally unsustainable. ### Real engineering value beyond the model If your product genuinely adds substantial engineering value, fine tuned models, custom retrieval, multi step orchestration, domain specific tooling, then pass through pricing leaves money on the table. The customer is getting more than the underlying model alone provides. Pricing only at the model rate underprices the actual value delivered. ### Significant infrastructure cost beyond the model Many AI products have non trivial infrastructure costs beyond the model API itself. Vector databases, fine tuning, evaluation pipelines, observability, support. If your product has these costs and you price at pass through, the model cost is covered but the rest of the operation is unfunded. The math eventually fails. ### Margin compression risk from provider rate changes Frontier model providers change rates over time, sometimes upward and sometimes downward. A pass through model passes the changes straight to the customer. When rates go down this is fine. When rates go up the customer sees a price increase that they did not opt into. This creates ongoing friction and erodes trust. A healthier margin gives you a buffer to absorb provider rate changes without immediately repricing. The customer relationship is more stable. ### Sales cycles that require margin to fund If your product needs sales motion, custom contracts, or significant onboarding, the cost of acquiring and supporting customers needs to be funded somewhere. Pass through pricing leaves no room for this. Either the subscription anchor needs to be fat enough to cover acquisition cost, or the model breaks at scale. For products with self serve adoption and minimal sales motion, this is less of an issue. For products with any meaningful go to market motion, pass through is hard to sustain. ## What healthy margin looks like For most AI products that are not in the specific situations where pass through earns its keep, a margin in the range of two to four times the underlying model cost is the sustainable shape. At two times, you have buffer for provider rate changes, support cost, and operational variance. The customer is paying a fair price for value above the raw model. The math works at scale. At three times, you have funding for engineering investment, sales motion, and the typical SaaS gross margin profile. The customer is still paying a defensible price for a product that is meaningfully more valuable than the raw model. At four times, you are positioned as a premium product. The customer is paying for substantial value beyond the model and you have margin to invest aggressively in the product. Above four times, you are charging premium pricing for AI capability. This works in some categories but you need to defend the value proposition continuously. Below two times, the variance in operational cost catches up with you. One bad month of higher than expected support load or one provider rate increase compresses your margin to zero. ## The mental model: cost basis plus value markup The cleanest way to think about AI product pricing is to separate the cost basis from the value markup. The cost basis is what you pay upstream. Model costs, infrastructure costs, vector store costs, support costs allocated per active customer. This is the floor. Pricing below this floor is unsustainable in the medium term. The value markup is what your product adds on top of the cost basis. The user interface, the integration, the specific workflow, the engineering value. This is the source of margin and the basis of pricing power. A pass through product has zero or low value markup. A premium product has high value markup. Most AI products land somewhere in the middle, with two to three times the cost basis being the sustainable range. When you reprice, look at both axes. Are your costs changing? Are you delivering more or less value? The right price moves with both. Customers accept price changes that reflect real changes in either direction. They resist arbitrary increases that have no underlying basis. ## What this means for your product If you are pricing an AI product right now, walk through the following. Calculate your true cost basis per active customer. Include model spend, infrastructure, support, and overhead. Pricing below this is unsustainable. Estimate the value markup your product delivers above the raw model. If the markup is genuinely small, you are in pass through territory and should price accordingly with a healthy subscription anchor. If the markup is substantial, you have margin room and should use it. Pick a multiplier on the cost basis that fits the markup. Two times is typical for products with modest markup. Three times for products with strong markup. Four times for premium positioning. Check that the resulting price fits the buyer's expectation for your category. Sophisticated developer audiences will balk at four times pricing for a thin wrapper. Less informed audiences will tolerate higher multiples for the same product. Plan for repricing. The underlying economics will move. Build the data and the operational ability to adjust prices when they need to change. ## How Dodo Payments fits Dodo Payments supports both pass through and margin based pricing models through subscriptions and usage based meters. You configure the subscription anchor and the overage rate, and the platform handles the billing, tax, and global compliance. The Merchant of Record model means you do not need to handle different tax rates or refund policies for different regions. For metering AI specifically, the platform offers ingestion blueprints for AI SDK, OpenAI, Anthropic, and other providers. The blueprints emit usage events automatically and feed the meters that drive your pricing. This is the same plumbing whether you charge pass through or marked up rates. For implementation reference see the [LLM ingestion blueprint](https://docs.dodopayments.com/developer-resources/ingestion-blueprints/llm) and the [usage based billing guide](https://docs.dodopayments.com/developer-resources/usage-based-billing-guide). This article gives you the pricing framework. The platform gives you the implementation surface. ## Closing thought Pass through pricing is the right answer in a small set of specific situations. For most AI products, a healthy margin on top of the underlying model cost is the sustainable shape. The buyer accepts it because the product is delivering value above the raw model. The seller can fund engineering, support, and growth. Both sides feel the price is fair. Claude Code is interesting because the parent company effectively absorbs the model cost internally, which makes the pass through positioning work in a way it would not for a third party product. Other AI products do not have that structural advantage and need to charge for the value they add. If you are building an AI product, do the math on your cost basis and your value markup. Price honestly above both. Resist the urge to underprice for adoption when the math does not support it, and resist the urge to overprice when sophisticated buyers will see through the markup. The right price is the one that reflects what your product actually does and what it actually costs to deliver. ## FAQ ### Is pass through pricing the same as cost plus pricing? Roughly yes. Cost plus pricing adds a fixed margin to the cost basis. Pass through is a near zero margin version of the same idea. Both contrast with value based pricing where the price reflects the customer benefit rather than the seller cost. ### How do I know if I have enough margin? Look at gross margin on the variable portion of your bill. If it is below two times your underlying cost, you have no buffer for variance. If it is above five times, you may be leaving money on the table or pricing yourself out of deals. The right range depends on your product but two to three times is typical for sustainable AI SaaS. ### What happens when provider rates change? With healthy margin you absorb modest changes without repricing. With pass through you reprice immediately and the customer feels the change. With aggressive margin compression you absorb everything and it eats your business. Plan for rate changes by building the right margin from the start. ### Should I publish my margin to customers? Most products do not publish the margin explicitly. They publish prices. Sophisticated customers can compute the margin from public rate cards. The honest pricing strategy is to charge what you can defend on value, not to hide the math. Customers respect transparent value based pricing more than opaque markups. ### Can I switch from pass through to healthier margin later? Possible but painful. Customers who signed up at pass through prices will resist repricing. The cleaner path is to start with a margin that is defensible and adjust within a narrow band over time. Starting with pass through and trying to move to healthy margin usually means losing the customers who signed up for the original deal. --- - [More AI articles](https://dodopayments.com/blogs/category/ai) - [All articles](https://dodopayments.com/blogs)