TechAILatest.newsTrackers

AI Model API Pricing Comparison, November 2026

The cheapest current AI model APIs in our tracker are Muse Ember (Meta) and Claude Haiku 5.5 (Anthropic) at $0.10 per million input tokens. The most expensive is GPT-6 Astra at $10.00. Below are list prices for 19 text and reasoning models, sorted from cheapest to most expensive input, with the cost of the same sample workload on each.

Updated 13 Nov 2026 from the AI model release tracker. Prices in USD per 1M tokens, standard tier, as published by each developer.

ModelDeveloperInput $/1MOutput $/1MContext10M in + 2M out
Muse EmberMeta0.100.501M tokens$2
Claude Haiku 5.5Anthropic0.10 (prompts up to 100k tokens); 0.50 above0.50 (prompts up to 100k tokens); 2.50 abovenot stated$2
MiMo-V2.6-Flash (309B total, 15B active)Xiaomi0.140.281M tokens$1.96
Inkling 2 Small (594B total, 17B active)Thinking Machines Lab0.301.00not stated$5
DeepSeek-V4.1-Flash-1017DeepSeek0.30 peak; 0.15 off-peak1.20 peak; 0.60 off-peak1M tokens$5.40
MiMo-V2.6-Pro (1.02T total, 42B active)Xiaomi0.435 (cache miss)0.871M tokens$6.09
Mistral Large 4 (about 1T total, 52B active)Mistral AI0.70 (0.07 cached)2.101M tokens$11.20
GPT-6.1 TerraOpenAI1.005.001,050,000 tokens$20
Gemini 4 HydrogenGoogle DeepMind1.507.501,048,576 tokens$30
Nano Banana 2.1Google DeepMind1.5030 (image output; 7.50 for text and thinking output); 0.0336 per 1K imagenot stated$75
Grok 4.7xAI2.00 (below 200,000 prompt tokens); 4.00 above6.00 (below 200,000 prompt tokens); 12.00 above500,000 tokens$32
Gemini 4 Carbon (general availability: paid Gemini API, AI Ultra)Google DeepMind2.00 until 30 November (4.00 from 1 December)10.00 until 30 November (20.00 from 1 December)not stated$40
Kimi K3.1 Pro (2.8T total, about 104B active)Moonshot AI2.0010.00About 1M tokens$40
Gemini 4 ArgonGoogle DeepMind2.00 (4.00 from 1 Dec 2026)10.00 (20.00 from 1 Dec 2026)1,048,576 tokens (unified input and output)$40
GPT-6.1 SolOpenAI2.0010.001,050,000 tokens$40
Claude Sonnet 5.5Anthropic2.0010.001M tokens$40
Claude Opus 5.5Anthropic4.0020.001M tokens$80
Claude Fable 5.2 (and Mythos 5.2)Anthropic8.0040.001M tokens$160
GPT-6 AstraOpenAI10.0050.001,050,000 tokens$200

Budget models under $1 per million input tokens

7 models cost less than $1 per million input tokens: Muse Ember (Meta), Claude Haiku 5.5 (Anthropic), MiMo-V2.6-Flash (309B total, 15B active) (Xiaomi), Inkling 2 Small (594B total, 17B active) (Thinking Machines Lab), DeepSeek-V4.1-Flash-1017 (DeepSeek), MiMo-V2.6-Pro (1.02T total, 42B active) (Xiaomi) and Mistral Large 4 (about 1T total, 52B active) (Mistral AI). Developers position most of them for high-volume agent steps, extraction and routing, often next to a larger model that plans the work.

Frontier models at $2 or more

9 models list $2 or more per million input tokens. On the sample workload of 10 million input and 2 million output tokens, the cheapest model in the table costs $1.96 (MiMo-V2.6-Flash (309B total, 15B active)) and the most expensive costs $200 (GPT-6 Astra).

How to read these prices

The numbers are list prices on each developer's own pricing page or release notes on the day we recorded them. Several models charge more above a prompt length, offer cheaper cached input, or have an introductory price with an end date; the price cells show those notes. Image, video and speech models are priced in other units and are left out here; they are in the full tracker. Open-weight models you can run yourself are listed in our open-weight models guide, and every release is covered in AI models news.

Questions people ask

Which AI model API is the cheapest in November 2026?
Muse Ember (Meta) and Claude Haiku 5.5 (Anthropic), at 0.10 per million input tokens, among the 19 text models in our tracker with a published list price. Check the table for output prices and long-prompt surcharges.
Which AI model API is the most expensive?
GPT-6 Astra from OpenAI, at $10.00 per million input tokens and $50.00 per million output tokens.
How do I compare prices across models?
Output tokens cost several times more than input tokens, so compare a realistic mix. The last column prices 10 million input and 2 million output tokens at list price. Cached input, batch discounts and long-prompt surcharges change the real bill; the price cells note them where the developer publishes them.