ABTO
Category: AI Infrastructure
Korean student-founded startup offering an AI gateway and live A/B testing service that compares LLM model cost against real user behavior (conversions, retention) instead of published benchmark scores. ABTO was founded in 2026. The company is led by 강동혁 (Kang Dong-hyuk). Based in Seoul, South Korea. Team size: 1-10. Latest round: Bootstrapped.
- Founded
- 2026
- Headquarters
- Seoul, South Korea
- Team size
- 1-10
Value proposition
Helps companies pick the most cost-effective AI model for each feature by measuring real user outcomes (payment conversion, retention, response rate) alongside AI call costs — rather than relying on vendor benchmark scores that don't reflect a specific service's users.
Products and solutions
AI Gateway (OpenAI SDK-compatible, base_url swap), LLM model/prompt A/B testing with traffic splitting (e.g., 10% canary), Client SDK for event collection and conversion analytics, dashboard for per-model performance vs cost, Free / Pro (₩99,000/30 days) / Enterprise tiers.
Unique value
Outcome-based (real user behavior + cost) model selection instead of benchmark-based; drop-in OpenAI SDK compatibility (only base_url and headers change); per-feature experimentation via x-abto-feature-id; agent-assisted metric instrumentation.
Target customer
Companies and developers running production LLM/AI features (AI resume services, AI study/error-note services, etc.) seeking to optimize AI spend against real business outcomes.
Industries served
AI/LLM application developers, SaaS, consumer AI services (education, HR/resume, content)
Technology advantage
Gateway-level traffic splitting with multi-stage fallback/failover so customer services are unaffected; statistical-significance-aware A/B testing; outcome-attached cost scoring; OpenAI SDK drop-in compatibility; agent (Claude Code/Codex) skill that auto-instruments success metrics.
How they differentiate
Unlike benchmark-based model selection or pure observability tools, ABTO ties model choice to real user conversion/retention behavior and cost simultaneously, per feature, with a drop-in gateway that requires no code rewrite.
Main competitors
Helicone, Langfuse, Braintrust (also LangSmith, PostHog in LLM observability/A-B testing space)
Key partnerships
Participant in AI·SW Maestro 17th cohort (Ministry of Science and ICT / IITP program), beta deployments with AI resume and AI study-note services.
Major milestones
Launched live A/B testing service (Sept 2026), featured in Platum, GeekNews, and promppy, developed through AI·SW Maestro 17th cohort.
Growth metrics
Beta case studies cited: AI resume service achieved 28% model cost reduction and 12% increase in paid conversion; AI error-note service achieved 41% model cost reduction and 18% increase in revisit rate.
Market positioning
Early-stage challenger in the LLM observability / AI gateway / model A/B testing space, differentiated by outcome-and-cost-based model selection; positioned for Korean and global AI application developers.
Geographic focus
South Korea (primary); global LLM developer tooling market
About 강동혁 (Kang Dong-hyuk)
Computer Science student at Korea University (고려대학교 컴퓨터학과); developed AI services in the Ministry of Science and ICT / IITP-run AI·SW Maestro 17th cohort program before founding ABTO.
Latest news about ABTO
More AI Infrastructure companies
Official website: https://abto.app/