# Gemini and Mistral Are Doubling AI Prices. Price on List Rates

> Mistral's launch sale ended and Gemini's list prices double on set dates. Model your AI margins on list rates, not on promotional pricing.

- Canonical: https://zeroraised.com/blog/gemini-mistral-ai-price-increases
- Published: 2026-10-28
- Author: ZeroRaised Team (https://zeroraised.com/blog/authors/zeroraised)
- Publisher: ZeroRaised (https://zeroraised.com/)
- Section: Growth
- Tags: artificial-intelligence, revenue, saas, bootstrapping

Mistral has ended the launch sale on Mistral Large 4, doubling its price, and Google's pricing page shows Gemini 4 Argon and Gemini 3.8 Flash doubling on fixed dates. The AI price war is not a one-way ratchet: prices are moving in both directions. If your unit economics use the promotional number, they are already out of date.

**TL;DR**

- Mistral Large 4's launch sale, which lasted about a week, ended, moving $0.68 / $0.07 / $2.09 to $1.36 / $0.14 / $4.18 per 1M tokens ([UsagePricing](https://www.usagepricing.com/blueprint/activity/mistral-ai-2026-10-22-large-4-sale-ends)).
- Gemini 4 Argon input goes from $2.00 to $4.00 and output from $10.00 to $20.00 per 1M tokens on 1 December 2026 ([Google](https://ai.google.dev/gemini-api/docs/pricing)).
- Gemini 3.8 Flash moves from $0.75 to $1.50 input and from $3.75 to $7.50 output on 1 January 2027 ([Google](https://ai.google.dev/gemini-api/docs/pricing)).
- OpenAI went the other way, cutting GPT-6 Sol and Luna prices by 50% ([OpenAI](https://openai.com/index/introducing-gpt-6-sol-and-luna/)).
- Price on list rates, date every assumption, keep a fallback model and re-run comparisons when a provider changes its page.

**Disclosure:** ZeroRaised publishes this analysis and is a launch site for bootstrapped products. Facts come from the linked sources as checked on 28 Oct 2026; ZeroRaised has no commercial relationship with the companies named.

**List-rate pricing** is the standard public price a provider charges outside launch discounts, promotions and credit deals, and it is the number your unit economics should use.

## What exactly changed at Mistral and Google?

Mistral Large 4 is open-weight, with 52B active and 1.05T total parameters and a 1M context window, and its model page now lists $1.36 input, $0.14 cached input and $4.18 output per 1M tokens ([Mistral](https://docs.mistral.ai/models/mistral-large-4)). Those are double the launch prices of $0.68, $0.07 and $2.09 ([UsagePricing](https://www.usagepricing.com/blueprint/activity/mistral-ai-2026-10-22-large-4-sale-ends)).

Google's increase is scheduled rather than complete. Gemini 4 Argon, described as Google's flagship reasoning model, charges $2.00 input and $10.00 output per 1M tokens through 30 November 2026, then $4.00 and $20.00 from 1 December 2026; context caching moves from $0.10 to $0.20 ([Google](https://ai.google.dev/gemini-api/docs/pricing)). Output includes thinking tokens, Argon is not available on the Free tier, and every token is billed ([Google](https://ai.google.dev/gemini-api/docs/pricing)).

Gemini 3.8 Flash holds $0.75 input and $3.75 output through 31 December 2026, then moves to $1.50 and $7.50 on 1 January 2027 ([Google](https://ai.google.dev/gemini-api/docs/pricing)). The paid tier still gets a 50% cost reduction through the Batch API ([Google](https://ai.google.dev/gemini-api/docs/pricing)). That discount applies to the list rate, so a doubling still doubles the batch bill.

## Why does a short sale window matter more than the headline?

Because teams benchmark during promotions and then build a product around the result. The tracker's summary is blunt: "Any workload benchmarked on Large 4 during the sale doubles in cost at the same volume" ([UsagePricing](https://www.usagepricing.com/blueprint/activity/mistral-ai-2026-10-22-large-4-sale-ends)).

That is the trap. A launch sale is a customer acquisition tool, not a durable input price. If you set your own subscription price during the sale week, you priced a cost that no longer exists.

Gemini is friendlier: the increase is published in advance, so you have weeks to act.

## How do you build a list-rate cost model?

1. **Date every number.** Put the source page and the date you read it next to each cost. A cost without a date is an assumption wearing a suit.
2. **Use the highest listed rate.** If a price is scheduled to double, model the doubled rate: for Gemini 4 Argon, $4.00 input and $20.00 output per 1M tokens.
3. **Count thinking tokens.** Argon bills output, including thinking tokens, and every token is billed on the paid tier with no Free tier option ([Google](https://ai.google.dev/gemini-api/docs/pricing)). A model that thinks longer can cost more even at the same visible output length.
4. **Split cached and uncached input.** Mistral prices cached input at $0.14 against $1.36 for fresh input ([Mistral](https://docs.mistral.ai/models/mistral-large-4)). Caching changes the number a lot, so model both cases.
5. **Apply discounts after list rates.** Batch discounts and credits reduce a list rate. They should never replace it in the base case.
6. **Add a pass-through clause.** If you resell AI work, write the contract so provider price changes can be passed through with notice. If you sell fixed-price subscriptions, say how a material cost increase is handled.
7. **Re-run the comparison when the page changes,** not once a year.

## What does the OpenAI cut tell you?

OpenAI cut GPT-6 Sol to $2.00 input and $10.00 output per 1M tokens from $4.00 and $20.00 on GPT-5.6 Sol, and GPT-6 Luna to $0.10 and $0.50 from $0.20 and $1.20, describing the change as "50% cheaper" ([OpenAI](https://openai.com/index/introducing-gpt-6-sol-and-luna/)).

So the market is not drifting in one direction. Mistral's price has doubled and Google's is scheduled to, while OpenAI cuts. Assume costs only fall and the increases catch you; assume they only rise and you overpay. The boring correct posture: plan on the current list rate, date it, and keep switching cheap. Our guide to [reducing LLM costs](https://zeroraised.com/blog/how-to-reduce-llm-costs) covers the mechanics, and [how to choose an LLM API](https://zeroraised.com/blog/how-to-choose-an-llm-api) covers the criteria beyond price.

## Which model should you keep as a fallback?

Every production workflow should name a second model that can do the job at acceptable quality. For tasks where Large 4 fits, UsagePricing lists Mistral Large 3 at $0.5 / $1.5 and Medium 3.5 at $1.5 / $7.5 per 1M tokens ([UsagePricing](https://www.usagepricing.com/blueprint/activity/mistral-ai-2026-10-22-large-4-sale-ends)). Those are different models, so a fallback is only real if you have tested it.

That is why price changes and evals belong together: when a model doubles, re-run your test set against the alternative and compare pass rates at the new price. Our guide to [evaluating LLM outputs](https://zeroraised.com/blog/how-to-evaluate-llm-outputs) shows how to keep that set small. For open-weight options, [open-weight vs closed AI models](https://zeroraised.com/blog/open-weight-vs-closed-ai-models) covers the trade-offs, and self-hosting moves cost from tokens to operations.

## What this means for bootstrapped founders

1. **Reprice on list rates.** If your price was set on a promotional rate, raise it or cut scope before the promo ends.
2. **Show the date.** Add a "prices checked on" line to your internal cost sheet and to any public pricing claim.
3. **Keep a margin buffer.** A 2x input cost should hurt, not break you. Keep enough margin to absorb one model price move.
4. **Sell outcomes, not token counts.** Pair fixed-price plans with usage limits or fair-use terms; our guide to [pricing an AI product](https://zeroraised.com/blog/how-to-price-an-ai-product) has the patterns.
5. **Watch the whole market.** The [OpenAI and Anthropic discount war](https://zeroraised.com/blog/openai-anthropic-discount-war) is the other side of the same story: list prices are a moving target in both directions.
6. **Keep switching cheap.** Abstract the model call behind one interface, log token usage per feature, and know which features are price sensitive.

## Where ZeroRaised fits

ZeroRaised is a launch and discovery site for products built without institutional venture capital or accelerator equity. If you have a product people can try, you can [submit it here](https://zeroraised.com/submit). The [newsletter](https://zeroraised.com/newsletter) carries new analysis, and the [pricing page](https://zeroraised.com/pricing) explains the listing options.

A launch page is a good place to state how your product is priced and what happens if model costs change.

## Sources

- [Google Gemini API pricing](https://ai.google.dev/gemini-api/docs/pricing)
- [Mistral Large 4 model page](https://docs.mistral.ai/models/mistral-large-4)
- [UsagePricing: Mistral Large 4 sale ends](https://www.usagepricing.com/blueprint/activity/mistral-ai-2026-10-22-large-4-sale-ends)
- [OpenAI: introducing GPT-6 Sol and Luna](https://openai.com/index/introducing-gpt-6-sol-and-luna/)

## Frequently asked questions

### Did Mistral really double the price of Large 4?

Yes. Mistral Large 4 launched on a sale at $0.68 input, $0.07 cached input and $2.09 output per 1M tokens. The sale ended after about a week and the list price is now $1.36, $0.14 and $4.18, which is double the launch rate. Mistral documents the current price on its model page, and UsagePricing tracked the sale ending on 22 October 2026.

### When do the Gemini price increases start?

Gemini 4 Argon keeps its current paid-tier standard price through 30 November 2026 and doubles on 1 December 2026: input from $2.00 to $4.00 and output, including thinking tokens, from $10.00 to $20.00 per 1M tokens. Gemini 3.8 Flash doubles on 1 January 2027, with input moving from $0.75 to $1.50 and output from $3.75 to $7.50.

### Why should I use list rates instead of promotional prices?

A promotional price is temporary and often short. If you set your own price and margin on a sale rate, every customer you win at that margin becomes a loss when the sale ends. List rates are the number the provider can charge without a discount, so they are the safer basis for unit economics and for promises you make to customers.

### Does this mean AI prices are rising overall?

No. Prices are moving in both directions. OpenAI has also cut GPT-6 Sol to $2 input and $10 output from $4 and $20, and GPT-6 Luna to $0.10 and $0.50 from $0.20 and $1.20, which it describes as 50% cheaper. The lesson is not that AI is getting more expensive, it is that any single price can move and your plan should survive the move.

### What is the cheapest way to protect my margin?

Use cheaper models where the task allows, send batch work through discounted batch pricing, cache prompts when the provider supports it, and keep a tested fallback model. Then re-run your cost comparison at list prices whenever a provider changes its page. Do not assume the model you benchmarked last quarter still has the same price.