Oxlo.ai bets AI teams want flat-rate inference, not token bills; PH #2

Oxlo.ai was founded in Dubai in December 2025 by Barath Kanna, an AI-infrastructure and blockchain engineer who had led a 20-plus-engineer distributed organization at Trigan and founded the edge-AI startup Cyborg Network, with co-founder Megha Varshini Tamilarasan. Their bet was that token-based inference billing was the wrong mental model for the agent era: long-context prompts, RAG pipelines and agentic tool calls make token counts spike unpredictably, so they built a flat-fee inference API instead.

OxAPIs exposes 45+ open-source and frontier models — Kimi K2.6, DeepSeek, Llama 3.3, Qwen 3, Mistral, Whisper and more — behind a single OpenAI-compatible API, charging a flat rate per request rather than per token, so a 100-token prompt costs the same as a 50,000-token one. The platform promises zero data retention, no training on prompts, unlimited agentic tool calls, secure failover and an enterprise guarantee of 15% off a team's existing inference bill up to $20,000/month.

Oxlo launched on Product Hunt in late June 2026 and took #2 on the June 25 daily board with 486 upvotes. EveryDev's snapshot of oxlo.ai reports 700+ active users, 737M+ tokens processed and 15 employees, and Kanna's Stackforce profile (accessed 2026-09-02) says the platform reached roughly 3,500 users across 112 countries within about eight weeks, incubated by MCIT Qatar and an NVIDIA Inception member.

The open question is margin: flat per-request pricing on models that Oxlo itself buys per token could work as a loss leader for long-context workloads or squeeze the company as usage grows. An OxCompute GPU tier is listed as coming soon, pointing at a strategy of owning more of the stack to make the flat-rate economics hold.