Positioning — CalcLLM

Product A Production LLM Costs   Product B Dev Tool Costs   Platform Eng Lead Primary ICP   Startup CTO Secondary ICP

The inference cost barbell. Product A Every pricing tier dropped — GPT-4 launched at $30/$60 per MTok (Mar 2023); GPT-4.1 now delivers comparable capability at $2/$8 (−93%). Commodity models hit near-zero: GPT-4.1 nano and Gemini 2.5 Flash-Lite at $0.10/$0.40, DeepSeek V4 Flash at $0.14/$0.28. But new premium “max-effort” tiers appeared at the top: GPT-5.5 Pro at $30/$180, o3-pro at $20/$80 — creating a 750x input / 4,500x output price spread between cheapest and most expensive models (Epoch AI, a16z LLMflation). Meanwhile, frontier models are more expensive (GPT-5.5 Pro at $30/$180, Opus at $15/$75), and agentic workflows multiply consumption 5–30x per task. Companies are spending anyway — an expensive bet to avoid being left behind — but no one can trace the ROI. Enterprise GenAI spending surged 3.2x in one year ($11.5B → $37B) even as per-token costs fell (Menlo Ventures 2025). 85% of organizations misestimate AI costs by >10%. Product B The same dynamic is hitting dev tools: GitHub Copilot going usage-based June 2026 ($10–$39/mo in credits, then per-token per model); Claude Code averaging $150–250/dev/month at Opus rates; Cursor and Windsurf at $20–200/mo per seat. Costs shifting from predictable per-seat to opaque usage-based with zero native cost controls.

Tooling is shallow. Product A 17 competitors analyzed. All show what you spent. None show why the bill is 3x what token pricing implies. Zero products automate detection of retry overhead, cache miss rates, or context waste. Product B No product aggregates dev tool costs across GitHub Copilot, Claude Code, Cursor, and Windsurf into a single view.

Competitors are exiting. Product A Helicone acquired by Mintlify (March 2026), focus shifted to Mintlify integration. Portkey being acquired by Palo Alto Networks. Two major standalone players gone in three months.

Mid-market gap. Product A Product B CloudZero ($119M) and Finout ($85M) serve enterprise. StackSpend ($19/mo) and AI Cost Board ($9.99/mo) are shallow. Nobody serves platform eng leads at 50–500 person companies across both production API and dev tool costs.

Two cost domains, zero bridges. Product A Product B Production API costs and dev tool costs are managed in separate systems with no cross-visibility. Platform eng leads track LLM API spend in one dashboard and dev tool licenses in procurement spreadsheets. No product connects them. The team that bridges both owns the full AI cost picture.

Sources: Epoch AI price index, a16z LLMflation, Menlo Ventures 2025, Gartner March 2026, Zylo 2026, competitive-analysis.md (17 competitors), GitHub Copilot pricing, DeepSeek pricing

Product A thesis Product A LLM costs are fundamentally opaque in a way cloud costs never were. Cloud resources can be tagged; API calls cannot. The team that builds the diagnostic layer — explaining why the bill is higher than expected, not just what was spent — will own the category.

Post-launch costs exceed estimates by 30–60% from hidden multipliers alone: retry storms (2–5x), cache misses, context waste, provider billing quirks. The gap between sticker price and invoice is invisible until someone decomposes it.

Product B thesis Product B Dev tool costs are a black box of per-seat charges and usage overages. The team that aggregates and attributes them — showing per-engineer, per-tool, per-team cost across GitHub Copilot, Claude Code, Cursor, and Windsurf — wins. This is an aggregation bet, not a diagnostic depth bet. The value is cross-tool visibility that no single vendor provides.

Product A Platform eng leads don’t search for “cost intelligence” or “FinOps for AI.” They search with questions: “why is my LLM bill so high,” “OpenAI API cost breakdown by team,” “Anthropic billing attribution.”

Product B Startup CTOs search: “AI coding tool costs,” “Claude Code pricing,” “Copilot vs Cursor cost comparison,” “how much should I spend on AI dev tools.” The homepage should match the question, not the category.

Not demographics. Structural position. They sit at the intersection of three teams that don’t talk to each other about AI costs:

• Engineering sees per-request behavior but not aggregate cost.
• Finance sees the aggregate invoice but can’t attribute it.
• Product decides which features to build but has no visibility into per-feature AI cost.

The platform lead is accountable to all three. When the bill spikes, it lands on their desk. Scope spans both products: they manage the LLM API gateway and the dev tool procurement for engineering.

Psychographic: the Pragmatic Adopter. Within this structural role, our sharpest buyer believes AI probably delivers transformative value. They’re actively experimenting — $10–50K/month committed — but need to answer “is this working?” before scaling further. They’re making a proactive investment decision, not reacting to a surprise bill. CalcLLM converts their gut feel into data.

Why not VP Eng? Needs SOC 2 + SSO + 3-9 month procurement. Why not FinOps? Cloud framework mindset.

Not a specialized buyer. They ARE engineering + finance + product. At 10–50 people, the CTO owns the credit card, picks the tools, and explains the burn rate to the board.

• Discovery: Twitter/HN, not Gartner. Peer recommendations, not vendor demos.
• Decision speed: 1–3 days from discovery to payment. No procurement cycle.
• Payment: Personal or company credit card. Self-serve. No invoicing.
• Core need: Per-feature unit economics for board decks. “Our AI features cost $X per user per month.”

Evidence: AI Vyuh has paying startup customers at $50–$2K/mo. Validates WTP for this ICP.

Startup CTO → Platform Eng Lead As a startup grows past ~50 people, the CTO hires a platform eng lead. The same CalcLLM account graduates from CTO-driven to platform-team-driven. Product B (dev tool costs) is the landing product; Product A (production API costs) is the expansion product. One account, different buyer stages.

Not Observability No request tracing, no prompt management, no eval, no latency metrics. Langfuse ($4.5M) and Braintrust ($80M Series B) own observability. Cost is a checkbox in their platforms. CalcLLM goes deep on cost because it doesn’t go wide on observability.

Read-Only First Starts with read-only billing API pull. No proxy, no SDK, no code changes. This removes the adoption blocker that killed Helicone’s growth. Optional SDK/telemetry unlocks deeper multiplier detection (retry attribution, cache miss rates) in later tiers — honest about the tension: read-only limits diagnostic depth, but the progressive path lets trust build before asking for integration.

Not Optimization No model routing, no prompt compression, no spending limits. Cohrint optimizes tokens. AI Vyuh recommends downgrades. CalcLLM diagnoses: “retries cost you $2K/month.” Facts are easier to sell than prescriptions. Optimization can follow once trust is established.

Not Enterprise FinOps No cloud resource management, no tagless allocation, no enterprise procurement. CloudZero ($119M) and Finout ($85M) own enterprise FinOps. CalcLLM is AI-native, not cloud-first with AI bolted on.

Not a Billing System No payments, no invoices, no chargeback accounting. Provides data that feeds existing finance workflows. Building finance integrations would consume engineering time without advancing diagnostic value.

1. Product A Provider APIs expose enough data. Can read-only APIs detect ≥2 of 5 hidden multiplier categories without request-level telemetry? Risk: Medium. Plan for progressive integration.
2. Platform Eng Lead Platform eng leads own the budget. Do they self-identify as budget owner for LLM costs, or does it sit with VP Eng / CTO / FinOps? Risk: Low. Multiple data points confirm.
3. Product A Diagnostic framing resonates. “Why is your bill 3x?” converts better than “see what you spent.” Risk: Medium. A/B testable on landing page.
4. Platform Eng Lead Mid-market WTP at $25–60K/year. Platform eng leads will pay $100–200/seat/mo for cost intelligence. Risk: Medium. Indirect signals positive, needs direct validation.
5. Product A Solo founder can ship Product A to PMF. API integration + multiplier detection + attribution buildable in 6–12 months by one person. Risk: High. Anti-positions exist to keep scope manageable.
6. Product B Solo founder can ship Product B to PMF. Dev tool aggregation + per-engineer attribution buildable in 3–6 months by one person. Risk: Medium. Simpler integration surface (vendor billing APIs), but sequencing against Product A is an open question (see Q7).
7. Product B Dev tool aggregation has lasting value. Falsifiable within 6 months as GitHub ships budget controls (June 2026). If each vendor provides adequate native cost controls, the aggregation layer loses value. Risk: Medium. Claude Code and Cursor have zero cost controls today; vendor incentives favor usage, not cost transparency.
8. Startup CTO Startup CTOs will pay for cost intelligence. Sub-50-person companies will pay $50–200/mo for dev tool cost visibility. Risk: Medium. AI Vyuh has paying startup customers. Counter-risk: CTOs may prefer spreadsheets at this scale.