Last updated: 2026-06-29
Touchmark (YC S26) scores AI outputs on quality evals and adjusts billing per result. TypeScript SDK, 10-line integration, built for AI SaaS product teams.
Touchmark is a Y Combinator S26 startup that prices AI outputs by quality, not by token count. Its TypeScript SDK integrates in under 10 lines of code, scores every output on evals including code quality, instruction following, and user satisfaction, and adjusts the bill per result. Poor outputs cost less or are refunded. Touchmark also publishes industry AI indices used in Fortune 500 procurement and financial instruments.
Touchmark is a quality-based AI pricing platform built by Y Combinator S26 founders Ilia Bolgov and Roman Yanushevskyi. The company was founded in 2026 with a single thesis: almost all AI products are sold by token count, with a correct answer costing exactly the same as a wrong one. Touchmark solves this by running automated evaluations on every AI output a product generates, then adjusting the final price charged to the customer based on those quality scores. Poor outputs are automatically discounted or refunded; high-quality outputs can command a slightly higher price. The result is a billing system where customers pay for what they actually received. The integration is a TypeScript SDK (@touchmark/sdk) that drops into any existing AI product with under 10 lines of code. Developers keep all existing logic, add a session open call, tag each model output with a base price, and then apply the delta adjustments Touchmark returns as quality scores settle. The default eval suite covers at least 5 dimensions: code quality, instruction following, user satisfaction, hallucination rate, and deterministic signals. All scores normalize to a [0, 1] scale, making third-party and custom evals plug-compatible with the default suite. Every priced output gets a public session page at sessions.touchmark.ai showing the score breakdown, the calculation, and a plain-language reason a non-technical user can read. Beyond the billing layer, Touchmark publishes three industry AI quality indices: the Pharma AI Index, the Code AI Index, and the Legal AI Index. These are updated daily, built from the combination of relevant evals weighted by industry usage, and are citable in enterprise procurement decisions. Fortune 500 procurement teams reference these indices when evaluating AI vendor quality. The quality timeseries Touchmark produces also backs financial instruments: forwards let buyers lock in AI quality at a future price, call options give buyers the right to exercise if outputs clear a quality threshold, and exchange-listed futures settle in cash against the eval timeseries. Sovereign wealth funds have taken long positions on the Pharma AI Index. Touchmark is designed for AI SaaS companies that sell AI-generated outputs to paying customers, whether that is code review, content generation, research, or any domain where quality varies and customers notice the difference. The SDK works with any model or agentic app, from a raw API call to a full multi-step agent. Pricing for the Touchmark platform itself is custom and contact-based as of mid-2026, reflecting the early-stage nature of the product and the negotiated billing adjustments between Touchmark and each AI vendor. The founding team brings a mix of quantitative finance experience (Citadel Securities) and product management at consumer fintech (Revolut), which explains the unusual combination of eval infrastructure and financial derivative instruments that defines the platform. Touchmark was accepted into Y Combinator's Summer 2026 batch, one of the first S26 companies to address AI output pricing rather than AI product creation itself.
Custom pricing, contact-based. No published tiers as of June 2026. Pricing is usage-based, negotiated per AI vendor based on eval volume and billing delta volume. No free tier.
Touchmark is a Y Combinator S26 startup, founded in 2026 by Ilia Bolgov and Roman Yanushevskyi, that prices AI outputs by quality rather than by token count. The platform runs automated evaluations on every AI output a product generates, covering dimensions like code quality, instruction following, user satisfaction, and hallucination rate, then adjusts the price charged to the end customer based on those scores. High-quality outputs may cost a little more; poor outputs cost less or are refunded automatically, with the revenue impact averaging to zero across all users on the same evaluation. Developers integrate via the @touchmark/sdk TypeScript package, which records sessions and emits priced events without changing existing application logic. Every priced output gets a public session page at sessions.touchmark.ai showing the quality scores and a plain-language explanation. Touchmark also publishes industry AI quality indices (Pharma, Code, Legal) used in Fortune 500 procurement and as reference rates for financial instruments including futures and options.
Touchmark does not publish standard pricing tiers on its website as of June 2026, as it is an early-stage YC S26 startup operating in beta with custom pricing agreements for each AI vendor partner. The platform's pricing model is itself usage-based: AI product companies pay Touchmark based on the volume of outputs evaluated and the billing delta volume generated per session. There are no published free tiers, starter plans, or enterprise price points available publicly. Companies interested in using Touchmark need to contact the team directly to negotiate pricing based on expected eval and session volume. Touchmark's fee to the AI vendor is separate from the quality-adjusted price the end customer pays, which adjusts per output. As a YC S26 company with 2 founders, pricing terms are likely flexible for early design partners.
Touchmark's core feature is quality-adjusted billing: a default eval suite covering at least 5 dimensions (code quality, instruction following, user satisfaction, hallucination rate, and deterministic signals) scores every AI output and adjusts the final charge accordingly, with all scores normalized to a [0, 1] scale. The TypeScript SDK @touchmark/sdk is designed for 10-line integration, recording sessions, emitting model_output and code diff events, and returning quality-adjusted price deltas per session. Every priced output gets a public session page at sessions.touchmark.ai showing the score breakdown, the calculation, and a plain-language explanation that end users can read. Touchmark also publishes three daily-updated industry AI quality indices (Pharma AI Index, Code AI Index, Legal AI Index) citable in Fortune 500 procurement decisions. Financial derivative products including forwards, call options, and exchange-listed futures allow enterprise buyers and sovereign wealth funds to hedge or gain exposure to AI output quality risk over time. Third-party and custom evals can plug into the same [0, 1] scale alongside the default suite.
Touchmark does not offer a free plan as of June 2026. It is a B2B platform for AI product companies that want quality-based billing for their own customers, so access requires direct contact with the Touchmark team rather than a self-serve signup flow. There is no published free trial or sandbox tier listed on the website. Early-stage AI companies accepted into the beta program may receive discounted or deferred pricing as design partners, which is standard practice for YC S26 companies in the early-access phase. The closest thing to a free experience is the public session viewer at sessions.touchmark.ai, where anyone can inspect a quality-adjusted price explanation for a given session ID. Touchmark is designed for commercial AI product builders with paying customers rather than for individual developers experimenting with LLM pricing on personal projects.
Touchmark's closest alternatives are LLM observability and eval platforms, though none offer the same quality-adjusted billing mechanism as their primary feature. Braintrust is the most mature alternative for eval automation with CI/CD pipeline blocking and prompt iteration tools; choose Braintrust if you need a proven, closed-source eval platform for internal engineering quality gates today. Langfuse is an open-source option (MIT licensed, acquired by ClickHouse in early 2026) with a free self-hosted plan; choose Langfuse if you need full data ownership and are willing to build your own pricing logic on top of its tracing data. LangSmith (from LangChain) offers eval and tracing tightly integrated with the LangChain framework; choose LangSmith if you are already using LangChain and want native integration without additional infra. Arize Phoenix provides open-source observability with OpenTelemetry support for teams that prefer OTEL-native tooling. None of these competitors offer Touchmark's financial instruments (futures, options, forwards) or its published AI quality indices used in enterprise procurement.
Touchmark is best for founders and engineering teams at AI SaaS companies that sell AI-generated outputs to paying customers and want to charge based on output quality rather than flat subscriptions or token usage. The ideal user is a startup or mid-size company building a product like an AI code review tool, an AI writing assistant, or an AI research agent, where output quality varies noticeably and customers have pushed back on paying the same price for poor outputs as great ones. Enterprise AI procurement teams that need a standardized third-party benchmark for comparing vendor output quality against competitors will also benefit from Touchmark's published indices. The financial instruments (futures, options on AI quality indices) are aimed at institutional investors and sovereign wealth funds, not at individual developers. Touchmark is not suited to solo developers building personal projects, non-technical teams, or companies that do not yet have a commercial AI product with paying customers.
Getting started with Touchmark requires contacting the team directly, since there is no public self-serve signup as of June 2026. Once accepted into the beta, the integration follows a 5-step pattern using the @touchmark/sdk npm package. First, install the package with npm install @touchmark/sdk. Second, create one client instance using your api_key. Third, open a session at the start of each user interaction. Fourth, emit a model_output event and tag it with the base_price_usd you would normally charge. Fifth, apply the quality-adjusted delta Touchmark returns to your customer billing system as eval scores settle. The full documentation including the quickstart guide, API reference, event types, and eval customization options is available at docs.touchmark.ai. Custom evals can be added to the default suite on the same [0, 1] scale to match domain-specific quality criteria for specialized use cases.
Touchmark and Braintrust both run evals on AI outputs, but they serve fundamentally different goals. Braintrust is an eval and observability platform for internal engineering workflows: teams use it to catch regressions in CI/CD pipelines, iterate on prompts, and trace production LLM calls, with a published free tier for teams up to a usage limit and enterprise plans for larger organizations. Touchmark is a customer-facing billing layer: it adjusts what your end customers pay based on the quality of outputs they received, not what your engineers see internally. Braintrust has been on the market since 2023 with a mature product, published pricing, and a large base of AI engineering users, while Touchmark is a YC S26 company (2026) with 2 founders, no published pricing, and an earlier-stage but more differentiated product. Choose Braintrust if you need internal eval automation, prompt management, and a proven track record for an engineering team today. Choose Touchmark if you are an AI SaaS company that wants to implement outcome-based customer billing and differentiate on quality-guaranteed pricing, and are willing to work with an early-stage vendor.