# Culpa

> Culpa is a local-first LLM cost, margin, and forecast ledger. It names the
> exact conversation that spent the money, tags it to the feature and the
> customer, and forecasts the next bill. Your prompts and responses never
> leave your infrastructure. Culpa counts calls to run your plan.

Guides hub: https://getculpa.com/guides
Machine-readable catalogue: https://getculpa.com/api/pages
Every page below has a plain-text mirror at {url}/llms.txt

## How to read the cost figures

Every cost figure on this site carries a confidence label and a source.
`calculated` shows its arithmetic. `estimated` is modelled and always a
range. `provider-reported` comes from a real provider export and appears
only on teardown pages.

## howto (1)

- [How to Size an LLM Cost Plan Before You Pick a Tier](https://getculpa.com/how-to-size-an-llm-cost-plan): Size a metered plan from tokens, not requests. Your tier moves when prompts grow even if traffic holds flat. The sizing arithmetic, worked.

## integration (11)

- [Track LLM Cost on Anthropic, Where Usage Arrives in Two Parts](https://getculpa.com/track-llm-cost-anthropic): Anthropic sends streamed usage across two events and charges to create a cache. What Culpa records, and where it deliberately takes a floor.
- [Track LLM Cost on Gemini: Two Doors, Different Numbers](https://getculpa.com/track-llm-cost-gemini): Gemini can be reached through its native API or an OpenAI-compatible endpoint. The two carry usage in different fields, so check which yours uses.
- [Track LLM Cost on Groq When a Model Has No Price Yet](https://getculpa.com/track-llm-cost-groq): Open-weight model lineups move fast, so a price book lags. What Culpa records for a model it has no rate for, and why a visible zero beats a guess.
- [Track LLM Cost in LangChain, and What the Adapter Can See](https://getculpa.com/track-llm-cost-langchain): Culpa's LangChain adapter is a callback handler that reports the framework's own token counts. An honest floor, and how to measure what it misses.
- [Track LLM Cost in LiteLLM: The Highest-Fidelity Adapter](https://getculpa.com/track-llm-cost-litellm): LiteLLM's response is already openai-shaped, so Culpa's adapter passes it through nearly verbatim. It also doubles as an import format for history.
- [Track LLM Cost for Ollama, Where No Invoice Ever Arrives](https://getculpa.com/track-llm-cost-ollama): A local model sends no bill, so its cost is invisible rather than absent. How to price Ollama traffic against what the same tokens would have cost.
- [Track LLM Cost on OpenAI, and the Streaming Usage Trap](https://getculpa.com/track-llm-cost-openai): A streamed OpenAI call carries no token usage unless you ask for it. What that leaves in your ledger, and the one flag that turns it back on.
- [Track LLM Cost on OpenRouter: Price What Answered](https://getculpa.com/track-llm-cost-openrouter): A router can serve a different model than you requested. Culpa prices the model that actually answered, which is the only figure that matches the bill.
- [Track LLM Cost in Python Without Sending Your API Key](https://getculpa.com/track-llm-cost-python): Culpa's Python SDK reports the raw provider response to a local server. Your API key never travels, only a fingerprint, and capture adds no tokens.
- [Track LLM Cost in TypeScript When Your App Runs in the Cloud](https://getculpa.com/track-llm-cost-typescript): Culpa binds to localhost, so a cloud-hosted TypeScript app has no route home. What that leaves unattributed, and how a relay closes it.
- [Track LLM Cost in the Vercel AI SDK, and the Streaming Gap](https://getculpa.com/track-llm-cost-vercel-ai-sdk): Culpa's Vercel AI SDK middleware wraps generate calls. Streaming is out of scope for now, so a chat product captures a fraction of its own spend.

## margin (8)

- [AI COGS Calculator: What Serving One Customer Really Costs](https://getculpa.com/ai-cogs-calculator): An AI COGS calculator totals model spend, retries and cache behaviour per customer. What belongs in cost of goods sold, and what doesn't.
- [AI SaaS Gross Margin Calculator: The Four Inputs It Needs](https://getculpa.com/ai-saas-gross-margin-calculator): An AI SaaS gross margin calculator needs four inputs, and most teams guess three. Margin computed blended, per tier and per customer, with math.
- [AI Usage Cap Calculator: Where to Set the Ceiling](https://getculpa.com/ai-usage-cap-calculator): An AI usage cap calculator sets the ceiling from your 90th-percentile account, not the average. How to pick a cap that binds the tail and nobody else.
- [How to Price an AI SaaS Product Without Losing Money](https://getculpa.com/how-to-price-an-ai-saas-product): How to price an AI SaaS product: start from your 90th-percentile customer cost, not the average, and cap what sits beyond it. Worked math inside.
- [How to Price a Credit-Based AI Plan So a Credit Covers Cost](https://getculpa.com/price-a-credit-based-ai-plan): A credit-based AI plan only works when one credit maps to bounded spend. Why flat credits break on uneven actions, with worked math.
- [How to Price a Per-Seat AI Product When Seats Differ](https://getculpa.com/price-a-per-seat-ai-product): Per-seat pricing decouples revenue from usage until seats stop looking alike. Why dormant seats fund power seats, and what activation does.
- [How to Price a Usage-Based AI Product on a Real Markup](https://getculpa.com/price-a-usage-based-ai-product): Usage-based pricing passes model cost through, so the markup is the whole decision. Why a markup on accepted spend under-recovers, with math.
- [How to Price an Unlimited AI Plan Without Going Broke](https://getculpa.com/price-an-unlimited-ai-plan): Pricing an unlimited AI plan means pricing against a maximum nobody has measured. What unlimited really costs, and the safer alternatives.

## metric (19)

- [What Is a Deep Trace? One Conversation, Down to the Prompt](https://getculpa.com/what-is-a-deep-trace): A deep trace opens one conversation down to the prompt. That's a scarce forensic unit, so aim it at the top of the cost distribution, not at alerts.
- [What Is a Tracked Call? The Unit a Metered Plan Counts](https://getculpa.com/what-is-a-tracked-call): A tracked call is one metered event, and a big call costs more than a small one. The meter counts work rather than requests. The arithmetic, shown.
- [What Is a Blended Rate? Averaging LLM Cost Across Models](https://getculpa.com/what-is-blended-rate): A blended rate is your average cost per token across every model you run. Useful for forecasting, dangerous for optimisation. Worked example inside.
- [What Is LLM Cost Attribution? Linking Spend to Its Cause](https://getculpa.com/what-is-cost-attribution): LLM cost attribution links each dollar to the call, conversation, feature and customer behind it. Why it has to happen at request time, not in reporting.
- [What Is a Cost Cross-Check? Calculated Against Reported](https://getculpa.com/what-is-cost-cross-check): A cost cross-check compares your own calculated figure against the provider's reported one, with a tolerance small enough to catch a wrong rate.
- [What Is a Cost Floor? A Number That Admits What It Missed](https://getculpa.com/what-is-cost-floor): A cost floor is the lowest defensible figure for a call, recorded when part of the input is absent and flagged rather than estimated into completeness.
- [What Is Cost per Agent? The Whole Run, Not the First Call](https://getculpa.com/what-is-cost-per-agent): Cost per agent is the total spend for one autonomous run, including sub-calls, tool steps and retries. Why one run is a tree, with worked math.
- [What Is Cost per Agent Step? Why Step 20 Costs More](https://getculpa.com/what-is-cost-per-agent-step): Cost per agent step rises as a run goes on, because each step resends the transcript before it. Why doubling steps more than doubles spend.
- [What Is Cost per Conversation? Definition and Worked Math](https://getculpa.com/what-is-cost-per-conversation): Cost per conversation is the total model spend for one full exchange, retries and tool calls included. Definition, the arithmetic, and how to measure it.
- [What Is Cost per Customer? Why It Isn't Cost per User](https://getculpa.com/what-is-cost-per-customer): Cost per customer is model spend attributed to a paying account, not to one person inside it. Why seat growth breaks the per-user view, with math.
- [What Is Cost per Feature? Tagging LLM Spend to Product Surfaces](https://getculpa.com/what-is-cost-per-feature): Cost per feature is model spend grouped by the product surface that triggered it. Why provider data can't produce it, and how to tag calls so it can.
- [What Is Cost per Prompt Version? Pricing Every Edit](https://getculpa.com/what-is-cost-per-prompt-version): Cost per prompt version splits model spend by which revision served the call. Why a prompt edit is a pricing decision, with worked math.
- [What Is Cost per User for an AI Product? Median vs P90](https://getculpa.com/what-is-cost-per-user): Cost per user is model spend divided by active users, and the average lies. Why the 90th percentile sets your pricing floor, with worked math.
- [What Is Effective Cost? The Rate You Actually Pay per Token](https://getculpa.com/what-is-effective-cost): Effective cost is your real blended rate per token after caching, retries and failed calls. Why it always runs above the list price, with worked math.
- [What Is Forecast Accuracy? Two Numbers, Never One](https://getculpa.com/what-is-forecast-accuracy): Forecast accuracy needs both a median error and whether reality landed inside the band. Either one alone can make a useless forecast look good.
- [What Is Gross Margin per Customer for AI Products?](https://getculpa.com/what-is-gross-margin-per-customer): Gross margin per customer is revenue minus what that customer costs you in model spend. How to compute it per account and why averages hide losses.
- [What Is an LLM Burn Forecast? Predicting Next Month's Bill](https://getculpa.com/what-is-llm-burn-forecast): An LLM burn forecast projects model spend forward from your own usage curve. What makes one credible, and the three inputs teams usually get wrong.
- [What Is Prompt Cache Hit Rate? And When Caching Costs More](https://getculpa.com/what-is-prompt-cache-hit-rate): Prompt cache hit rate is the share of input tokens billed at the cached rate. Why a low rate wastes money and a high write rate can cost more.
- [What Is Token Accounting? Counting LLM Spend Without Floats](https://getculpa.com/what-is-token-accounting): Token accounting is tracking input, output and cached tokens per call and pricing them exactly. Why rounding and float math corrupt LLM cost data.

## model_compare (6)

- [Gemini 2.5 Flash vs Llama 3.3 70B Cost: Where They Cross](https://getculpa.com/gemini-2-5-flash-vs-llama-3-3-70b-versatile-cost): Gemini 2.5 Flash wins on retrieval work and Llama 3.3 70B wins on chat. They cross at an output-to-input ratio of 0.17. Worked math inside.
- [Gemini 2.5 Pro vs Gemini 3.5 Flash Cost: The Crossover](https://getculpa.com/gemini-2-5-pro-vs-gemini-3-5-flash-cost): Gemini 2.5 Pro is cheaper than Gemini 3.5 Flash on input-heavy work. The two cross at an output-to-input ratio of 0.25. The arithmetic, shown.
- [Gemini 3.1 Flash-Lite vs Llama 3.3 70B Cost: A Crossover You Reach](https://getculpa.com/gemini-3-1-flash-lite-vs-llama-3-3-70b-versatile-cost): These two cross at an output-to-input ratio of 0.48, which sits inside the range real features occupy. Routing per feature beats picking one model.
- [GPT-5.4 Mini vs Kimi K2 Cost: The Cheaper Input Rate Loses](https://getculpa.com/gpt-5-4-mini-vs-kimi-k2-instruct-0905-cost): GPT-5.4 mini's input rate sits a quarter below Kimi K2's and it still costs more on most workloads. The two cross at a ratio of 0.17.
- [GPT-5.4 Nano vs Llama 3.3 70B Cost: A Crossover at 0.85](https://getculpa.com/gpt-5-4-nano-vs-llama-3-3-70b-versatile-cost): GPT-5.4 nano beats Llama 3.3 70B on cost until output nearly equals input. The crossover sits at 0.85, which almost no workload reaches.
- [GPT-5.6 Luna vs Llama 3.3 70B Cost: The Crossover Moves](https://getculpa.com/gpt-5-6-luna-vs-llama-3-3-70b-versatile-cost): These two cross at 0.95 on short prompts and at 0.19 on long ones, so the cheaper model flips on prompt length rather than on token ratio.

## model_pricing (22)

- [Claude Fable 5 Pricing: $10 In, $50 Out, Same as Mythos 5](https://getculpa.com/claude-fable-5-pricing): Claude Fable 5 pricing is $10.00 per million input and $50.00 output, matched by Claude Mythos 5. The top Anthropic tier and what a cache write costs.
- [Claude Haiku 4.5 Pricing: $1 In, $5 Out per Million Tokens](https://getculpa.com/claude-haiku-4-5-pricing): Claude Haiku 4.5 pricing is $1.00 per million input and $5.00 output, with cached reads at $0.10. The million-call tier where small errors get large.
- [Claude Opus 4.8 Pricing: $5 In, $25 Out, Same as 4.5 to Opus 5](https://getculpa.com/claude-opus-4-8-pricing): Claude Opus 4.8 pricing is $5.00 per million input and $25.00 output, identical to Opus 4.5, 4.6, 4.7 and Opus 5. Why the version upgrade is free.
- [Claude Sonnet 5 Pricing: $2 In, $10 Out Until 2026-08-31](https://getculpa.com/claude-sonnet-5-pricing): Claude Sonnet 5 pricing is $2.00 per million input and $10.00 output on an introductory rate that ends 2026-08-31, rising 50% to $3.00 and $15.00.
- [Gemini 2.5 Flash Pricing: $0.30 In, $2.50 Out per Million](https://getculpa.com/gemini-2-5-flash-pricing): Gemini 2.5 Flash pricing is $0.30 per million input and $2.50 output, an 8.3x spread. Why answer format is the whole optimisation here.
- [Gemini 2.5 Pro Pricing: $1.25 In, $10.00 Out per Million](https://getculpa.com/gemini-2-5-pro-pricing): Gemini 2.5 Pro pricing is $1.25 per million input and $10.00 output, the widest output spread in this price book at 8x. Long-context math inside.
- [Gemini 3.1 Flash-Lite Pricing: The Guardrail Tax](https://getculpa.com/gemini-3-1-flash-lite-pricing): Gemini 3.1 Flash-Lite pricing is $0.25 per million input and $1.50 output. Put it in front of every request and it earns out above 2.2% diversion.
- [Gemini 3.1 Pro Preview Pricing: A Rate With an Expiry](https://getculpa.com/gemini-3-1-pro-preview-pricing): Gemini 3.1 Pro Preview pricing is $2.00 per million input and $12.00 output. Preview rates move, so a forecast pinned to one ages fastest.
- [Gemini 3.5 Flash Pricing: The Newer Flash Costs 5x More](https://getculpa.com/gemini-3-5-flash-pricing): Gemini 3.5 Flash pricing is $1.50 per million input and $9.00 output, five times Gemini 2.5 Flash on input. A version bump here multiplies the bill.
- [GPT-5.4 mini Pricing: $0.75 In, $4.50 Out per Million](https://getculpa.com/gpt-5-4-mini-pricing): GPT-5.4 mini pricing is $0.75 per million input and $4.50 output, about a third of GPT-5.4. Where it saves money and where volume eats the saving.
- [GPT-5.4 Nano Pricing: The Bottom of a 150x Ladder](https://getculpa.com/gpt-5-4-nano-pricing): GPT-5.4 nano pricing is $0.20 per million input and $1.25 output, 150x below the Pro tier. The widest cost decision sits inside one vendor.
- [GPT-5.4 Pricing: $2.50 In, $15.00 Out per Million Tokens](https://getculpa.com/gpt-5-4-pricing): GPT-5.4 pricing is $2.50 per million input and $15.00 output, verified 2026-07-02. Output bills 6x input, with worked monthly math.
- [GPT-5.4 Pro Pricing: $30 In, $180 Out, and No Cached Rate](https://getculpa.com/gpt-5-4-pro-pricing): GPT-5.4 Pro and GPT-5.5 Pro both bill $30 per million input and $180 output, with no cached rate. What that means for repeated prompts.
- [GPT-5.5 Pricing: $5.00 In, $30.00 Out, Same as GPT-5.6 Sol](https://getculpa.com/gpt-5-5-pricing): GPT-5.5 pricing is $5.00 per million input and $30.00 output, exactly double GPT-5.4. When the premium earns itself, with the break-even math.
- [GPT-5.6 Luna Pricing: $0.20 In, $1.20 Out per Million](https://getculpa.com/gpt-5-6-luna-pricing): GPT-5.6 Luna pricing is $0.20 per million input and $1.20 output, the cheapest OpenAI tier. Why volume, not rate, decides what it costs you.
- [GPT-5.6 Terra Pricing: $2.00 In, $12.00 Out per Million](https://getculpa.com/gpt-5-6-terra-pricing): GPT-5.6 Terra pricing is $2.00 per million input and $12.00 output, undercutting GPT-5.4 by 20% on both sides. The migration math, worked.
- [Llama 3.1 8B Pricing on Groq: Cheap Enough to Stop Counting](https://getculpa.com/llama-3-1-8b-instant-pricing): Llama 3.1 8B pricing on Groq is $0.05 per million input and $0.08 output. At that rate a call rounds to nothing and volume becomes the whole bill.
- [Llama 3.3 70B Pricing on Groq: The Flattest Ratio in the Book](https://getculpa.com/llama-3-3-70b-versatile-pricing): Llama 3.3 70B pricing on Groq is $0.59 per million input and $0.79 output, a 1.34x ratio. Why shortening answers barely moves this bill.
- [Kimi K2 Pricing on Groq: The Cached Rate Nobody Switches On](https://getculpa.com/moonshotai-kimi-k2-instruct-0905-pricing): Kimi K2 pricing on Groq is $1.00 per million input and $3.00 output, with a cached read at half the input rate. On agent traffic that's a quarter off.
- [GPT-OSS-120B Pricing on Groq: Exactly Twice the 20B](https://getculpa.com/openai-gpt-oss-120b-pricing): GPT-OSS-120B pricing on Groq is $0.15 per million input and $0.60 output, exactly double the 20B at every rate. So the only question is quality.
- [GPT-OSS-20B Pricing on Groq: A Cache Worth Half, Not 90%](https://getculpa.com/openai-gpt-oss-20b-pricing): GPT-OSS-20B pricing on Groq is $0.075 per million input and $0.30 output. Its cached rate is about half input, not the 90% discount you expect.
- [Qwen3.6 27B Pricing on Groq: Five-to-One Output, No Cache](https://getculpa.com/qwen-qwen3-6-27b-pricing): Qwen3.6 27B pricing on Groq is $0.60 per million input and $3.00 output. A reasoning-tuned model on a 5x spread with no cached rate to fall back on.

## optimization (3)

- [Batch Processing Cost Optimization: 50%, and How to Lose It](https://getculpa.com/batch-processing-cost-optimization): OpenAI, Anthropic and Gemini all bill batch at half price. Caching stacks on top of it on a best-effort basis, which can invert the saving.
- [Context Window Cost Optimization: 200k Is a Cliff, Not a Slope](https://getculpa.com/context-window-cost-optimization): One token past Gemini's 200k boundary doubles the input rate and lifts the output rate by half. The whole prompt reprices, not the excess.
- [Prompt Cache Cost Optimization: The Second Call Decides It](https://getculpa.com/prompt-cache-cost-optimization): A cache write costs 1.25x input and a read costs 0.1x, so caching pays from the second call inside the window. Volume isn't the variable.

## problem (21)

- [Agent Fan Out Cost: Why One Request Becomes Thirty Calls](https://getculpa.com/agent-fan-out-cost): An agent turns one user request into many model calls, and cost grows with steps rather than requests. How to price agent fan out before it ships.
- [Anthropic Bill Spike: Cache Writes Are Usually the Cause](https://getculpa.com/anthropic-bill-spike): An Anthropic bill spike with flat traffic usually means cache writes. A write costs more than plain input, so a cache nobody reads is a surcharge.
- [Context Creep Cost: When Your Prompt Grows and Nobody Notices](https://getculpa.com/context-creep-cost): Context creep is the slow growth of prompt size across releases. It raises cost per call with no traffic change. How to measure and cap it.
- [Failed Call Cost: You Pay for Output You Never Read](https://getculpa.com/failed-call-cost): A failed call bills for every token generated before your code gave up. Timeouts, cancelled streams and rejected output all cost. The math, shown.
- [Free Tier Abuse: What One Automated Signup Really Costs](https://getculpa.com/free-tier-abuse): Free tier abuse is scripted signups consuming a full allowance at maximum cost per call. One can cost what a hundred genuine trials do.
- [Gemini Bill Spike: The 200k Context Tier Nobody Priced](https://getculpa.com/gemini-bill-spike): A Gemini bill spike with flat traffic often means prompts crossed 200,000 tokens, where the long-context rate roughly doubles. Worked math inside.
- [Input Token Creep: Four Percent a Month Is 60% a Year](https://getculpa.com/input-token-creep): Input token creep is a small monthly rise across your whole workload. No single change looks big, and twelve of them compound to 60%. Math inside.
- [LLM Cost Spike After Launch: What Actually Drives It](https://getculpa.com/llm-cost-spike-after-launch): A launch spike is rarely just traffic. Free-tier users, curiosity sessions and unbounded retries drive it. How to forecast and cap the next one.
- [LLM Eval Job Cost: The Line Item Nobody Watches](https://getculpa.com/llm-eval-job-cost): Evaluation runs bill like production but serve no customer. Why eval spend hides, how to price a run, and how to cut it without losing coverage.
- [LLM Model Mix Drift: The Bill Rises With No Code Change](https://getculpa.com/llm-model-mix-drift): Model mix drift is traffic quietly shifting toward pricier models. Volume holds, the bill climbs. How to detect it and price the shift.
- [LLM Retry Loop Cost: Why Failed Calls Still Bill You](https://getculpa.com/llm-retry-loop-cost): A retry loop bills every attempt, including the ones that produced nothing you kept. How to price your retry waste and cap it, with worked math.
- [Model Upgrade Cost Increase: Same Workload, New Rate](https://getculpa.com/model-upgrade-cost-increase): A model upgrade cost increase is the same workload at a different rate, and it can happen without you changing anything. Worked math inside.
- [OpenRouter Bill Spike: Your Route Changed, Not Your Code](https://getculpa.com/openrouter-bill-spike): An OpenRouter bill spike on flat traffic usually means a different upstream provider served your calls. Same model, about 10% more per token.
- [Output Token Creep: The Rise Input Monitoring Can't See](https://getculpa.com/output-token-creep): Output token creep raises cost while input stays flat, so prompt-size monitoring shows nothing. Output bills 5x to 8x input. Worked math inside.
- [Power User AI Cost: Why Your Heaviest Account Is Heavy](https://getculpa.com/power-user-ai-cost): Power user AI cost is usually one workflow, not general enthusiasm. How to decompose your heaviest account and find the loop inside it.
- [Prompt Cache Miss Cost: What a Broken Cache Really Costs](https://getculpa.com/prompt-cache-miss-cost): A prompt cache miss bills your repeated tokens at full rate, and on some providers a cache write costs more than input. The math, and the fix.
- [Prompt Version Cost Increase: Finding the Edit Behind It](https://getculpa.com/prompt-version-cost-increase): A prompt version cost increase shows up as a step change on one date with flat volume. How to bisect it back to the edit, with worked math.
- [RAG Context Cost: Top-K Is a Pricing Decision](https://getculpa.com/rag-context-cost): RAG context cost is top-k multiplied by chunk size on every call. Retrieval settings set your input bill more than your prompt does.
- [Tool Definition Token Cost: You Pay for Every Unused Tool](https://getculpa.com/tool-definition-token-cost): Tool schemas are input tokens billed on every call, whether the model uses them or not. How to price your tool block and trim it safely.
- [Unattributed LLM Spend: The Bucket That Quietly Grows](https://getculpa.com/unattributed-llm-spend): Unattributed spend is every dollar your breakdown can't assign to a feature or customer. Why it grows, and how to drive it toward zero.
- [Why Did My OpenAI Bill Go Up? The Four Causes, Ranked](https://getculpa.com/why-did-my-openai-bill-go-up): Your OpenAI bill went up for one of four reasons: more calls, longer calls, a pricier model, or retries. How to tell which one, with the math.

## provider_cost (5)

- [Anthropic API Cost: Every Model Bills Output at Exactly 5x](https://getculpa.com/anthropic-api-cost): Anthropic API cost runs $1.00 to $10.00 per million input, with every model at exactly 5x output, and it charges to create a cache as well as read one.
- [Gemini API Cost: The Widest Output Spreads, Up to 8.3x](https://getculpa.com/gemini-api-cost): Gemini API cost runs $0.25 to $2.00 per million input, with output spreads up to 8.3x. Why answer length is the lever on this provider.
- [Groq API Cost: Where Trimming Your Output Barely Helps](https://getculpa.com/groq-api-cost): Groq API cost runs $0.05 to $1.00 per million input, with output ratios as flat as 1.34x. Why the usual cost advice stops working here.
- [OpenAI API Cost: A 150x Spread From Nano to Pro](https://getculpa.com/openai-api-cost): OpenAI API cost runs from $0.20 to $30.00 per million input tokens, a 150x spread. Why model choice matters more here than on any other provider.
- [OpenRouter API Cost: One Model, Several Prices](https://getculpa.com/openrouter-api-cost): OpenRouter routes one model to several upstream providers at different prices. The same Opus call bills $5.00 or $5.50 depending on who served it.

## teardown (2)

- [Model Mix Cost Teardown: The Cheap Model Did More Work](https://getculpa.com/model-mix-cost-teardown): A real month, measured. The small model processed more input and twelve times more output than the frontier model beside it, for 55% of the cost.
- [Prompt Cache Cost Teardown: The Month Caching Cost Money](https://getculpa.com/prompt-cache-cost-teardown): A real month of Anthropic API usage, measured. The account wrote 2,810,172 tokens into caches and read 571,898 back, which is below break-even.