Official and OpenRouter prices checked daily
See what the same workload costs through a model provider or OpenRouter. Compare source-linked rates, include routing fees, and choose the channel that fits your stack.
Official direct pricing and OpenRouter pricing stay separate, so a cheaper route never gets mistaken for a provider price.
Jev 1.13
Jev is built for typed decisions rather than free-form prose. It returns scores and probabilities over explicit choices for review, routing, classification, and policy workflows. TypeSafe AI reports throughput up to 250K tokens per second and 1,200 RPM.
Put messages, records, or rules in state.
Define a choice, score, or true/false question.
Call POST /v1/systemone with jev-latest, then route by the answer and confidence.
Structured decisions and prose generation solve different jobs, so Jev is kept out of the cheapest text-generation ranking.
Cost calculator
Choose a model and workload once. Switch channels to see the price difference without changing your inputs.
Compare any two models or the same model through two channels. Edit the sample workload to match your application.
Stored sample, 2,000 input + 500 output tokens, 10,000 requests: A $125.00 / B $135.00. Enable JavaScript to edit and see the detailed comparison.
Links contain your model choices and workload numbers. Share only estimates you want others to see. Costs use stored rates; check each source date below before budgeting.
Price comparison
| Model | Official / 1M | OpenRouter / 1M | Output difference | Context | Sources |
|---|---|---|---|---|---|
OGPT-5.4OpenAI |
In $2.50Out $15.00 | In $2.50Out $15.00 | Same price | 1.05M | |
OGPT-5.4 miniOpenAI |
In $0.75Out $4.50 | In $0.75Out $4.50 | Same price | 400K | |
AClaude Opus 5Anthropic |
In $5.00Out $25.00 | In $5.00Out $25.00 | Same price | 1M | |
AClaude Sonnet 5Anthropic |
In $2.00Out $10.00 | In $2.00Out $10.00 | Same price | 1M | |
AClaude Haiku 4.5Anthropic |
In $1.00Out $5.00 | In $1.00Out $5.00 | Same price | 200K | |
GGemini 3.1 Pro PreviewGoogle |
In $2.00Out $12.00 | In $2.00Out $12.00 | Same price | 1.049M | |
GGemini 3.6 FlashGoogle |
In $1.50Out $7.50 | In $0.75Out $3.75 | 50% less via OpenRouter | 1.049M | |
DDeepSeek V4 ProDeepSeek |
In $0.435Out $0.87 | In $0.5946Out $1.1891 | 37% more via OpenRouter | 1M | |
MLlama 4 MaverickMeta |
No first-party API | In $0.1875Out $0.6525 | OpenRouter only | 1.049M | |
ZGLM 5.2Z.ai |
In $1.40Out $4.40 | In $0.5544Out $1.7424 | 60% less via OpenRouter | 1.049M | |
TAJev 1.13TypeSafe AI · structured-decision model |
In $0.042Out $0.00 | In $0.042Out $0.00 | Structured output; not prose-comparable | 66K |
No models match these filters.
Pricing guides
Use focused comparisons when you need a direct answer about the cheapest APIs, one provider, or the major model families.
Compare the lowest verified token rates and learn when the cheapest list price can still cost more in production.
View price ranking → Provider pricingSee Claude input, cached-input, output, OpenRouter, and example workload costs in one source-linked guide.
Compare Claude prices → Provider comparisonApply the same token workload to representative premium, balanced, and budget models from all three families.
Compare providers → Change trackingFollow verified official and OpenRouter price changes from a transparent baseline, without reconstructed history.
View price history →Explore focused routes: Google Gemini API pricing, OpenAI GPT API pricing, and the complete LLM API pricing hub.
Browse the complete LLM API pricing guide hub →How to use the data
Start with the workload, not the model name. Estimate fresh input tokens, repeated cached context, output length, and monthly request count. LLM API Prices applies those inputs consistently so you can compare first-party and OpenRouter routes without mixing their prices or fees.
A lower list price is only useful when the model still meets your requirements. Test response quality, tool support, latency, availability, context behavior, and data policies before moving production traffic.
LLM API prices are easiest to compare when every rate uses the same unit. This site normalizes published rates to USD per one million tokens, while keeping input, cached-input, and output prices visible as separate cost components. That makes a headline rate easier to connect to the way an application actually spends tokens.
Use the calculator for a quick estimate, then compare the assumptions with your own logs. A support bot with short prompts, a coding assistant with repeated context, and a batch summarization job can produce very different monthly totals on the same model. The result is a planning reference, not a quote: confirm current provider terms, rate limits, regional rules, and billing details before production use.
Price data also needs lifecycle context. A preview, retired, self-hosted, or region-specific model should not be treated as a universal production rate. Model pages separate lifecycle notes, context windows, official sources, OpenRouter mappings, and verification dates so you can understand what a number means before comparing it.
Jev 1.13 is TypeSafe AI's structured-decision model: it returns typed choices and probabilities instead of free-form text. Its official rate is $0.042 per million input tokens with free output. Compare it with decision workloads rather than treating its zero output rate as a text-generation price advantage.
When a source is unavailable or a change looks structurally suspicious, the catalog keeps the last verified record instead of silently publishing a guess. That distinction matters for teams using the site to plan budgets: a missing update should be visible as a verification issue, not disguised as a new price.
It calculates fresh input, cached input when a cache-read rate is available, generated output, cost per run, and monthly cost. For OpenRouter, you can optionally include the percentage charged when purchasing credits.
An official price is a verified first-party API rate linked to the model provider’s documentation. Open-weight models without a universal first-party hosted API are labeled instead of being assigned an invented price.
Taxes, regional premiums, batch tiers, priority processing, tool calls, storage, image or audio units, and special long-context rules are excluded unless the model note explicitly says otherwise.
Frequently asked questions
Providers usually charge separately for input tokens, cached input, and generated output. Multiply each token amount by its per-million rate, then add the components and monthly request volume.
Sometimes, but not always. Many OpenRouter inference rates match first-party prices, while some mapped routes differ. Credit purchase fees and operational features should be compared separately.
No. The calculator runs in your browser and does not ask for model-provider credentials, send prompts, or inspect your actual API usage.
Official sources and the OpenRouter model feed are checked daily. Last-known-good prices are preserved when a source fails, and suspicious official changes are held for review.
Comparison method
Official prices come from each model provider. Open-weight models without a first-party hosted API are labeled clearly.
OpenRouter prices come from its model API and remain a separate channel, with their own source and update time.
OpenRouter’s credit purchase fee is optional in the calculator, so it never disappears inside the token rate.