LLM API Prices started a verified baseline on Jul 29, 2026 for 47 official API routes and 50 OpenRouter routes. No real price change has been observed yet, so this page shows the baseline and methodology without inventing historical data.
What are the latest verified LLM price changes?
The change stream is intentionally strict. It does not treat a modified webpage, a missing response, a promotional banner, or a similarly named model as proof of a new token price. An event appears only after a verified rate differs from the last stored rate for that exact model and channel.
Tracking has started with a current-price baseline. This area will show the model, channel, affected token rate, old price, new price, percentage movement, date, and source after a real change passes validation.
What pricing baseline is being tracked?
The baseline is the first verified observation stored after history tracking began. It establishes a known starting point; it is not a claim that the price launched on that date. The table below shows a representative slice of the 50-model catalog. Prices are USD per one million tokens, shown as input / output. Cached-input prices remain stored in the underlying history even when the compact table does not display them.
| Model | Official in / out | OpenRouter in / out | Changes | Source checked |
|---|---|---|---|---|
| GPT-5.6 SolOpenAI | $5.00 / $30.00 | $5.00 / $30.00 | 0 | Jul 30, 2026 |
| GPT-5.6 TerraOpenAI | $2.50 / $15.00 | $1.25 / $7.50 | 0 | Jul 30, 2026 |
| GPT-5.6 LunaOpenAI | $1.00 / $6.00 | $0.50 / $3.00 | 0 | Jul 30, 2026 |
| GPT-5.4OpenAI | $2.50 / $15.00 | $2.50 / $15.00 | 0 | Jul 30, 2026 |
| GPT-5.4 miniOpenAI | $0.75 / $4.50 | $0.75 / $4.50 | 0 | Jul 30, 2026 |
| GPT-5.3 CodexOpenAI | $1.75 / $14.00 | $1.75 / $14.00 | 0 | Jul 30, 2026 |
| GPT-5.2OpenAI | $1.75 / $14.00 | $1.75 / $14.00 | 0 | Jul 30, 2026 |
| GPT-4.1OpenAI | $2.00 / $8.00 | $2.00 / $8.00 | 0 | Jul 30, 2026 |
| Claude Fable 5Anthropic | $10.00 / $50.00 | $10.00 / $50.00 | 0 | Jul 30, 2026 |
| Claude Opus 5Anthropic | $5.00 / $25.00 | $5.00 / $25.00 | 0 | Jul 30, 2026 |
| Claude Sonnet 5Anthropic | $2.00 / $10.00 | $2.00 / $10.00 | 0 | Jul 30, 2026 |
| Claude Opus 4.8Anthropic | $5.00 / $25.00 | $5.00 / $25.00 | 0 | Jul 30, 2026 |
| Claude Opus 4.7Anthropic | $5.00 / $25.00 | $5.00 / $25.00 | 0 | Jul 30, 2026 |
| Claude Opus 4.6Anthropic | $5.00 / $25.00 | $5.00 / $25.00 | 0 | Jul 30, 2026 |
| Claude Sonnet 4.6Anthropic | $3.00 / $15.00 | $3.00 / $15.00 | 0 | Jul 30, 2026 |
The full current catalog remains available on the 50-model comparison table. Each current price links to its provider or OpenRouter source.
How does daily price history tracking work?
After the scheduled source check, LLM API Prices compares each verified channel with its most recent history event. If input, cached-input, and output rates are identical, no event is added. If one or more fields changed, the append-only record stores the previous values, new values, percentage movement where it can be calculated, source URL, and observation time.
Official and OpenRouter data travel through separate update paths. Official pricing comes from first-party provider documentation and uses bounded parsers or manual review for suspicious changes. OpenRouter pricing comes from its public model feed and an explicit model-ID map. The history writer runs only after both current datasets have been refreshed, so it records the validated state rather than scraping a third version of the truth.
Why can a price change require manual review?
Provider price pages often contain several rates for the same model: standard, batch, regional, promotional, long-context, cache-write, cache-read, audio, image, or priority processing. A number can change on the page without changing the standard text-token rate used by the calculator. Large or structurally suspicious changes are therefore held before they can alter current pricing or history.
This protects the archive from false changes. A blocked source also keeps the last-known-good price instead of creating an artificial drop to zero. Deprecated models are marked with a lifecycle status rather than deleted, which allows future history pages to preserve the meaning of earlier observations.
How should you use historical API prices?
Price history is useful for budgeting, vendor reviews, and understanding whether a routing channel tracks a provider’s public list price. It should not be used as a capability benchmark. A cheaper model can still cost more in production if it needs longer outputs, more retries, more tool calls, or frequent escalation to a premium model.
For a decision today, start with the current comparison and run your own workload through the calculator. Use the history to see how pricing moved over time, then check latency, reliability, context behavior, data retention, regional availability, rate limits, and contract terms before moving production traffic.
LLM API price history questions
When did LLM API Prices start tracking price history?
The current baseline begins on Jul 29, 2026 using the source verification dates stored with each model. Earlier provider prices are not reconstructed or presented as observed history.
Does the price history combine official and OpenRouter prices?
No. Official provider prices and OpenRouter rates remain separate channels with separate source URLs and verification dates. A change in one channel never overwrites or implies a change in the other.
What counts as an LLM API price change?
A history event is added only when verified input, cached-input, or output pricing differs from the last stored event for the same model and channel. A daily check with unchanged pricing adds no event.
What happens when a pricing source cannot be checked?
The last-known-good event remains in the history. A failed, unavailable, or stale source does not create a new event and does not turn missing data into a zero price.
Why is this page initially excluded from search indexing?
A baseline is useful for transparency but is not yet a meaningful change archive. LLM API Prices keeps this page out of the Sitemap until it contains at least 3 verified change events.