CostLens — Know what AI will cost before your invoice does

Hey Product Hunt 👋

I'm Nishit, founder of Kipps.AI. We build AI voice, WhatsApp, and chat agents for businesses across India, the Middle East, and Southeast Asia. Every month, we're making thousands of API calls across multiple providers — and for the longest time, our AI cost estimates were embarrassingly wrong.

Not because we weren't tracking tokens. We were. We didn't know what we didn't know.

The problem that made us build this

Six months into production, our bills were consistently 3–4× our estimates. We started digging. Turns out the token price card is the tip of the iceberg:

• Reasoning models like o3 and DeepSeek R1 generate internal chain-of-thought tokens billed at full output rate — often 5–10× the visible response. Nothing on the pricing page tells you this upfront.

• GPT-4 tiles a 1024×1024 image into ~765 tokens before it reads your prompt. Send 10 images in a conversation, and your "cheap" chat call isn't cheap anymore.

• GPT-4o Realtime charges $100/1M audio input tokens vs $5/1M for text. Same model. 20× difference depending on how you send data.

• Batch API discounts give you 50% off on OpenAI, Anthropic, and Google — but only if your architecture is deliberately designed around it. Most aren't.

• Agentic loops that trigger web searches can quietly run up to $10–$35 per 1,000 calls, depending on the provider. One pipeline can fire dozens.

• STT providers round audio duration differently. AWS rounds up to 15 seconds. Deepgram bills per second. At scale, this gap is enormous, and none of the comparison tools calls it out.

We couldn't find a single tool that modelled all of this together. So we built one.

What CostLens is

A structured, open-source dataset and TypeScript SDK covering 97+ models across every AI modality:

• LLMs and multimodal (OpenAI, Anthropic, Google, Mistral, Meta, DeepSeek, xAI, Cohere, and more)

• STT — per-minute rates, billing granularity, streaming support

• TTS — per-character and per-audio-token pricing

• Realtime audio — text vs audio token split, the 20× gap most people miss

• Image generation — flat, per-megapixel, and token-based billing unified into one schema

• Video generation — per-second and per-clip rates

Beyond the raw numbers, the schema captures everything that actually drives cost: reasoning token billing, image tile counts, function call overhead, cached input discounts, batch multipliers, and web search surcharges.

Why open source

We're a team building on top of AI every day. The pricing data goes stale fast — new models launch, rates change, providers quietly update their billing. No single team can keep up. Community PRs are the only model that works here. MIT licensed, self-hostable, no accounts required.

Who this is for

If you're building anything with AI APIs and you've ever looked at your monthly invoice and thought, "how did it get this high?" — this is for you. Whether you're running a simple chatbot or a complex multi-provider agent pipeline, CostLens gives you the data layer to estimate before you commit.

Happy to answer anything — about the data model, missing providers, edge cases in billing we haven't captured yet, or how we handle modalities that don't fit the standard token schema. This is v1, and the gaps are real. Would love the community's help closing them.

— Nishit & the Kipps.AI team