Self-Host vs API Break-even

At what volume does self-hosting beat the API?

Running Llama 70B or Qwen 72B on rented GPUs. Factor in utilization, ops overhead, and the real costs most teams miss.

Pricing verified: 2026-07-28 H100 / A100 / cloud GPU rates
What this calculator does

At what volume does self-hosting (GPU cluster) beat vendor API pricing? Models breakeven across H100, A100, L40S, and major API providers.

Why use it
  • Self-host has high fixed costs (GPUs, ops) but low marginal cost per request — API is the opposite
  • Breakeven is workload-specific: tokens/request, GPU utilization, and ops overhead all swing it
  • Most "self-host saves 80%" claims ignore ops cost and assume 24/7 80%+ utilization — calculator forces honesty

New to this calculator? Start with the ⚡ Playground — a few sliders, instant ballpark. Then switch to the 🧮 Calculator for your exact number.

Two ways to use this: visualize in the Playground, then get your number in the Calculator.

▶Playground A quick, visual way to see which factors move your result the most. Open the playground → ▤Calculator Enter your real workload for a precise result you can apply to your own usage. Go to the calculator →
Self-Host Breakeven Playground
Drag the sliders to compare the options.
Self-host vs API — which is cheaper, and when?

API spend is what you pay today. Self-host trades it for fixed GPU + MLOps costs that only pay off at steady scale. GPU utilization fixed at 50%.

Cheaper option
—
Change the input sliders below to compare.

API
—
/ month
Self-host
—
/ month

💡Cheaper now is pure cost. Best fit weighs the catch: self-host’s GPU + MLOps is fixed, so it only wins with high, steady utilization.

Self-Host Breakeven Calculator

Enter your exact numbers for a precise result.
📊 Not sure of a value? Fields with a ▾ Typical pill offer broad industry ballparks (sourced typical ranges; Mid-2026: self-hosting breaks even only at sustained volume (analyses put the crossover at ~50M-250M tokens/mo vs premium APIs, and often never vs budget APIs). Ops labor is the hidden line: 10-20 hrs/mo minimum ($750-3K), a full MLOps FTE ~$145K/yr. Breakeven typically needs 60%+ GPU utilization.) so you can move forward now — your result gets more accurate as you replace them with your own measured numbers. Values marked * are rough estimates.
🎮 Interactive Guide & Calculator Playground →
Self-hosting trades a big fixed cost for cheaper inference. Enter your volume below to find the crossover point versus the API.
🎛 Your workload

Compare API inference against self-hosted on rented GPUs.

Hint: The pay-per-use API you'd compare self-hosting against.
Hint: Words sent in per request.
Hint: Words sent back per request.
Hint: Requests per day at your busiest.
🏗 Self-host setup
Hint: Which GPU instance you'd rent to self-host. * Throughput rates are approximations for 70B-class models. Smaller models 2-4x faster.
Hint: How busy the GPUs stay. Realistic production is 40-70%; idle time still costs you.
Hint: Engineer time to keep it running. ~$1,500-3,000/mo basic; more for production.
Hint: One GPU is a single point of failure; use 2+ for production reliability.

Results

-
-
-
📡 API-based
-
-
per month
🏗 Self-hosted
-
-
per month
⚖ Break-even volume
-
-
⚠ Real costs most teams miss
  • Model quality gap. Open-weight Llama 70B ≈ Claude Sonnet 3.5 / GPT-4o quality. Gap to current frontier (Opus 4.7, GPT-5.4) is real and closing slowly. Don't self-host a worse model to save money if quality matters.
  • Engineering time compounds. Model serving, autoscaling, monitoring, inference framework upgrades, security patching. Budget 0.25-0.5 FTE minimum for production.
  • No automatic new models. API vendors ship new models monthly. Self-hosting means you're stuck on whatever you deployed until someone ports + tests + redeploys.
  • Multi-region = multiply everything. One region works for demos. Production usually needs 2-3 regions for latency + DR. Multiply GPU count accordingly.
💡 Recommendations
    🖥 Monthly cost across GPU options

    At your volume + utilization. "Total" includes ops overhead + GPU count.

    GPU option $/hr Tokens/sec Raw GPU cost + Ops Total
    API cost calculator → Fine-tune instead? → Self-host feasibility audit →
    🎯 Use your Self-Host Breakeven results to…
    • ⚖️ Find your crossover — See the requests-per-day where self-hosting starts beating the API.
    • 🖥️ Size the GPU — Compare GPU instance types against your tokens-per-request to size the cluster.
    • 💰 Quantify the savings — See dollar savings at your real volume before committing to infrastructure.
    • 🔌 Integrate with your AI agents — MCP available for agentic workflow integration. Plug AICost.ai into your agents to surface real-time cost intelligence.

    Go deeper

    Our playbooks on cutting this number.

    🎓
    Fine-Tuning Cost
    Customize instead of self-host
    🎯
    Concentration Risk
    Why self-host = ultimate de-risk
    🧮
    API Cost Calculator
    The baseline comparison
    📊
    AI Unit Economics
    When self-host breaks even

    Need help using this calculator for your workloads?

    AICost.ai has 50+ calculators and playbooks. Schedule an AvatarVA meeting and we'll work through your real cost scenarios across AI & Cloud: visibility, cost reduction, optimization, forecasting and capacity planning, without sacrificing accuracy or performance.

    📅 Schedule an AvatarVA meeting →
    📖 Data sources & methodology 163 text models · 9 embeddings · 37 vision · 55 audio · 8 vector DBs across 10 vendor pages · last verified 2026-07-28

    Methodology

    • All prices are USD per 1 million tokens, current as of 2026-07-28.
    • Vendor-published values have no mark. Inferred/extrapolated values are marked with * and listed below.
    • Batch API discounts are 50% off standard rates across providers that offer Batch mode.
    • Prompt caching discounts vary by provider (typically 80-90% off cached input tokens).
    • Regional data-residency surcharges (Anthropic 1.1x, OpenAI 1.1x, Google regional tiers) are NOT included in base rates.
    • Long-context pricing tiers apply when input exceeds model threshold.
    • Embedding prices are input-only (no output tokens generated).

    Primary sources

    Last-verified date is the most recent successful daily snapshot (aicost_pricing_snapshots) or, when no snapshot exists yet, the latest successful crawler run (aicost_crawler_runs). 10 of 10 vendors are currently verified. Aggregator services (TokenCost, AI Pricing Guru, etc.) are not listed.

    Anthropic
    2026-07-28
    https://www.anthropic.com/pricing
    Daily snapshot since Sep 2023 · 631 days captured
    Anthropic Docs
    2026-07-28
    https://platform.claude.com/docs/en/about-claude/pricing
    Daily snapshot since Sep 2023 · 631 days captured
    OpenAI
    2026-07-28
    https://openai.com/api/pricing/
    Daily snapshot since Sep 2023 · 632 days captured
    Google AI
    2026-07-28
    https://ai.google.dev/gemini-api/docs/pricing
    Daily snapshot since Dec 2023 · 607 days captured
    Google Vertex
    2026-07-28
    https://cloud.google.com/vertex-ai/generative-ai/pricing
    Daily snapshot since Dec 2023 · 607 days captured
    DeepSeek
    2026-07-28
    https://api-docs.deepseek.com/quick_start/pricing
    Daily snapshot since May 2024 · 546 days captured
    xAI
    2026-07-28
    https://x.ai/api
    Daily snapshot since Nov 2024 · 464 days captured
    Mistral
    2026-07-28
    https://mistral.ai/pricing
    Daily snapshot since Dec 2023 · 605 days captured
    Cohere
    2026-07-28
    https://cohere.com/pricing
    Daily snapshot since Sep 2023 · 631 days captured
    Voyage AI
    2026-07-28
    https://docs.voyageai.com/docs/pricing

    Inferred values (marked with * in calculator tables)

    Derived from industry conventions, not directly published by the vendor. Typical conventions: cached input = 10% of base (90% off), Batch API = 50% of base (50% off).

    Vendor / Model Field Why it’s inferred
    Anthropic — Claude Sonnet 4.6 cachedInput Derived at 10% of input rate — Anthropic publishes 90% cache-hit discount on this tier.
    Anthropic — Claude Sonnet 4.5 cachedInput Derived at 10% of input rate; same 90% cache-hit convention as Sonnet 4.6.
    Anthropic — Claude Sonnet 4.5 batchInput Derived at 50% of standard input — Anthropic documents uniform 50% Batch discount.
    Anthropic — Claude Sonnet 4.5 batchOutput Derived at 50% of standard output — Anthropic documents uniform 50% Batch discount.
    Anthropic — Claude Haiku 4.5 cachedInput Derived at 10% of input rate — Anthropic 90% cache-hit discount convention.
    OpenAI — GPT-5.4 Mini cachedInput Derived at 10% of input — OpenAI documents automatic 90% discount on cache hits across GPT-5.x tier.
    OpenAI — GPT-5.4 Nano cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
    OpenAI — GPT-5.4 Nano batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
    OpenAI — GPT-5.4 Nano batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
    OpenAI — GPT-5.4 Pro cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
    OpenAI — GPT-5.4 Pro batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
    OpenAI — GPT-5.4 Pro batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
    OpenAI — GPT-5.2 cachedInput Derived at 10% of input; no residency uplift.
    OpenAI — GPT-5.2 batchInput Derived at 50% of input.
    OpenAI — GPT-5.2 batchOutput Derived at 50% of output.
    OpenAI — GPT-5 cachedInput Derived at 10% of input.
    OpenAI — GPT-5 batchInput Derived at 50% of input.
    OpenAI — GPT-5 batchOutput Derived at 50% of output.
    OpenAI — GPT-5.5 Pro cachedInput Derived at 10% of input — OpenAI does not publish a cached rate for *-pro models; using the family convention.
    OpenAI — GPT-5.5 Pro batchInput Derived at 50% of input.
    OpenAI — GPT-5.5 Pro batchOutput Derived at 50% of output.
    OpenAI — GPT-5.2 Pro cachedInput Derived at 10% of input — pro-tier convention.
    OpenAI — GPT-5.2 Pro batchInput Derived at 50% of input.
    OpenAI — GPT-5.2 Pro batchOutput Derived at 50% of output.
    OpenAI — GPT-5.1 batchInput Derived at 50% of input.
    OpenAI — GPT-5.1 batchOutput Derived at 50% of output.
    OpenAI — GPT-5 Pro batchInput Derived at 50% of input.
    OpenAI — GPT-5 Pro batchOutput Derived at 50% of output.
    OpenAI — GPT-5 Nano cachedInput Derived at 10% of input.
    OpenAI — GPT-5 Nano batchInput Derived at 50% of input.
    OpenAI — GPT-5 Nano batchOutput Derived at 50% of output.
    Google — Gemini 3 Flash cachedInput Derived at 10% of input — Google caching discount convention ~90%.
    Google — Gemini 3.1 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
    Google — Gemini 3.1 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
    Google — Gemini 3.1 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
    Google — Gemini 2.5 Pro cachedInput Derived at 10% of input.
    Google — Gemini 2.5 Flash cachedInput Derived at 10% of input.
    Google — Gemini 2.5 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
    Google — Gemini 2.5 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
    Google — Gemini 2.5 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
    Google — Gemini 2.0 Flash cachedInput Derived at 25% of input per Google 2.0 family caching rates.
    Google — Gemini 2.0 Flash batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
    Google — Gemini 2.0 Flash batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
    Google — Gemini 2.0 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
    Google — Gemini 2.0 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
    Google — Gemini 2.0 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
    xAI — Grok 4 (legacy) cachedInput Extrapolated at 25% of base.

    Pricing is cross-verified against the LiteLLM community registry when available. Daily snapshots are kept in aicost_pricing_snapshots; every change is logged to aicost_price_changelog with old & new values for full audit trail. Read the full methodology →