AI API Price Calendar for Q4 2026: Dates That Move Your Bill

Updated November 2, 2026 · first published November 2, 2026

Three published dates can change an AI API bill before the new year: Gemini 4 Argon doubles on December 1, GPT-5.6 Sol's promotional pricing is available at least through November 21, and Gemini 3.8 Flash doubles on January 1. Put each in your forecast as a dated step, not as a gradual trend.

Everything below was checked on 2 November 2026 against the providers' own pricing pages. Rates and dates can change; recheck them before you approve a budget.

Which price changes are scheduled?

For the Argon mechanics and a worked bill, read Gemini 4 Argon price doubles on December 1.

What is the exposure on the GPT-5.6 Sol date?

The wording matters. OpenAI's page lists GPT-5.6 Sol at $4.00 input, $0.40 cached input and $20.00 output per million tokens for prompts up to 272K tokens, and says promotional pricing is available at least through November 21. It does not name a post-promotion rate or say that the listed figures are the promotional ones. Do not invent an end-of-promo price. Instead, model two cases: the listed rate holds, and a higher rate applies, and set a trigger to re-forecast the day OpenAI publishes a change. For the full tier picture, see the GPT-5.6 pricing tier guide.

For context only, the same page lists GPT-6 Sol at $2/$10, GPT-6.1 Terra at $1/$5 and GPT-6 Luna at $0.10/$0.50. Those are different models, so a lower list price is a routing question to test, not a drop-in saving.

How do you put a dated step in a forecast?

1. Use a price table, not a single rate. Store each model's rate with an effective date. Compute each day's or week's cost from tokens x the rate in force on that day.
2. Hold volume constant, change price. Create a December scenario with November's token volumes and the new rates. The difference is pure price variance. Anything beyond that is usage. This is the method in the AI spend variance bridge.
3. Name the owner of each date. Someone should be responsible for confirming the rate on the provider page one week before the date.
4. Watch for contracts that override list price. Committed-spend deals can change what you actually pay; see committed spend discounts.
5. Decide the fallback in advance. If the cost of a model rises, which workloads move to a cheaper model, Batch, or a lower effort setting? Pre-agree the test (cost per accepted task) rather than deciding under invoice pressure.

Which prices are not scheduled to change?

Anthropic's Sonnet line is a useful counterexample. A planned Sonnet 5 increase was cancelled, and Sonnet 5 and 5.5 are listed at $2/$10; see the correction. A scheduled rise is a probability, not a certainty, so keep the dated step in your base case until the provider page changes, and keep a note of the source and date you last checked.

Scope and limits

This calendar lists only changes that a provider's own pricing page states. Third-party trackers report other moves, such as a Mistral Large 4 price change, that we could not confirm on Mistral's pricing pages as of 2 November, so they are not included. We will add them when a primary source states the rate and date.

Related

• Gemini 4 Argon price doubles on December 1
• GPT-5.6 pricing tier guide
• AI spend variance bridge
• Committed spend discounts

Want this applied to your stack? Bring the provider bills, gateway logs, and top workflows; we will map the cost drivers and savings path. Book a free audit →

Back to finopsllm.com