AI API prices, October 2026: same chatbot, a 100x gap

$13.95 and $1,395.

That's the monthly API bill for the same customer-support chatbot. The only thing that changed was the model, and the gap is 100x.

Between late September and early October, new models shipped from Claude, GPT and Grok, and the price sheets moved with them. So we re-checked five providers on their official pricing pages and priced one workload the same way across all of them.

※ Prices checked Oct 11, 2026 (KST). Standard tier, short prompts. Prices change often, so check the official page before you pay.

What's inside

The price board, per million tokens
Turning it into a monthly chatbot bill
Three things the price board doesn't show: caching, long prompts, tokenizers
New models this month
How to run your own numbers

1. The price board

Input / cached input / output, in USD per million tokens.

Standard prices for 13 models from five providers, from the official pages on 2026-10-11

Roughly three tiers:

Top tier: Claude Fable 5.1 and GPT-6 Astra, at $10 in and $50 out
Mid tier: Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 3.1 Pro (preview) and Grok 4.7, at $2 in and $6–12 out
Light tier: Claude Haiku 5.5 and GPT-6 Luna, at $0.10 in and $0.50 out

Gemini 3.8 Flash is on a promo price ($0.75 in, $3.75 out) through Dec 31, 2026. After that it doubles to $1.50 in and $7.50 out.

2. Turning it into a monthly bill

Per-token prices are hard to feel, so we fixed one workload:

2,000 requests a day for 30 days
1,500 input and 300 output tokens per request
half the input read from cache (system prompt and other repeated parts)

That comes to 45M regular input tokens, 45M cached input tokens and 18M output tokens a month.

Monthly bill for the same workload. No batch discount or free allowance applied.

GPT-6 Astra $1,395; Claude Fable 5.1 $1,361.25
Claude Opus 5.5 $549
Gemini 3.1 Pro $315
Claude Sonnet 5.5 and GPT-6.1 Sol $274.50
Grok 4.7 $220.50
Gemini 3.8 Flash $104.63 (about $209 once the promo ends)
Gemini 3.5 Flash-Lite $59.85
Workers AI GPT-OSS 120B $45; Gemma 4 26B $14.40
Claude Haiku 5.5 and GPT-6 Luna $13.95

3. Three things the price board doesn't show

Cache discounts differ. Cached input costs 5% of regular input (a 95% discount) on Claude Sonnet 5.5, Claude Opus 5.5 and GPT-6.1 Sol, and just 2.5% on Claude Fable 5.1. It's 10% (90% off) on Gemini, GPT-6 Astra and Luna, and Claude Haiku 5.5. On Grok 4.7 it's 25% (75% off). The more your prompts repeat, the more this matters.

GPT-6.1 Sol is here.

Upgraded with stronger agentic coding and computer use, near-Astra performance, and cached input at a 95% discount to standard input pricing.
— OpenAI Developers (@OpenAIDevs) 2026-09-30

Long prompts change the rate. Past a certain length, prices step up:

Claude Haiku 5.5: above 100K tokens, $0.50 in / $2.50 out (5x)
Gemini 3.1 Pro: above 200K, $4 / $18
Grok 4.7: above 200K, $4 / $12
GPT-6 Astra: above 272K, $20 / $75

Tokenizers differ. Anthropic notes that Claude 4.7 and later models count about 30% more tokens for the same text. The same per-token price can still mean a different bill.

4. New models this month

Claude Haiku 5.5 lists 90% below Haiku 4.5 ($1 in, $5 out). Anthropic says it costs "around 75% less to run" on average. The list-price gap and the real bill can be two different numbers.

Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.

On average, it costs around 75% less to run than Claude Haiku 4.5.
— Claude (@claudeai) 2026-10-08

Grok 4.7 launched at the same price as Grok 4.6.

Grok 4.7 is here.

It's a notable improvement over Grok 4.6 at the same price and speed.
— SpaceX Super Intelligence (@SpaceXSI) 2026-09-22

Gemini 3.8 Flash launched at the same promo price as 3.7 Flash.

Gemini 3.8 Flash is here⚡️

it's our most intelligent workhorse model, delivering significant improvements from 3.7 Flash across software engineering, agentic tasks, and critical, multi-step reasoning in specialized domains

available at the same introductory price as 3.7 Flash
— Google AI Studio (@GoogleAIStudio) 2026-09-03

One more: on Sept 30 (US time), Google introduced Gemini 4 Argon at an introductory $2 in and $10 out. It's rolling out to cyber defenders first, so we left it out of the comparison.

Introducing Gemini 4 Argon, our new frontier model, rolling out to cyber defenders starting today, and more widely as soon as possible. I am really excited by the progress we have made here. Argon is priced at $2 in and $10 out during introductory pricing!
— Logan Kilpatrick (@OfficialLoganK) 2026-10-01

5. How to run your own numbers

The numbers above come from one workload. Change the request count, token lengths or cache share, and the ranking can move.

Enter requests per day and input/output tokens per request.
Set the cache share to the part of your prompt that repeats (system prompt, documents).
If results can wait, try batch (50% off). Anthropic, Google and OpenAI list it.
If you send long documents, check whether long-prompt pricing kicks in.

The cast_otter AI API Cost Calculator covers all of these. The example above matches its "Support bot" preset.

Sources

Anthropic, Claude pricing (checked 2026-10-11): https://docs.claude.com/en/docs/about-claude/pricing
Google, Gemini API pricing (checked 2026-10-11): https://ai.google.dev/gemini-api/docs/pricing
OpenAI, API pricing (checked 2026-10-11): https://platform.openai.com/docs/pricing
xAI, models and pricing (checked 2026-10-11): https://docs.x.ai/docs/models
Cloudflare, Workers AI pricing (checked 2026-10-11): https://developers.cloudflare.com/workers-ai/platform/pricing/
Embedded posts are from the providers' official X accounts. Gemini 4 Argon pricing is from a post by Google's Logan Kilpatrick.
Monthly bills are cast_otter calculations (standard tier, short prompts, no batch, free allowance or tax).

Run these numbers yourselfAI API Cost CalculatorMonthly spend across 29 Claude, Gemini, GPT, Grok and Cloudflare models, with caching and batch discounts.

Share on XEverything on castotter.com is for information only and is not investment advice. We check numbers against public filings, but mistakes can happen. Your decisions are your own.