Kimi vs MiniMax vs GLM vs DeepSeek API pricing

Four Chinese AI labs, four price lists. This page puts Moonshot's Kimi, MiniMax, Z.ai's GLM and DeepSeek side by side, in USD per 1M tokens, with each price read on the maker's own pricing page. Together AI's prices for the same models are in a separate table and labelled as reseller prices. If we couldn't verify a figure, we left it out and say so.

Last verified: · every price re-read that day on platform.kimi.ai, docs.z.ai, platform.minimax.io, api-docs.deepseek.com and together.ai. They can change any of them. For the other model makers, see cheapest AI models.

First-party API prices

First-party API prices in USD per 1M tokens, from each maker's own pricing page, checked 30 Oct 2026. Sorted by blended cost (3 input : 1 output). DeepSeek rows show the peak rate.
ModelInput /1MOutput /1MCached input /1MBlended 3:1ContextNotesSource · checked
GLM-5.3-FlashZ.ai (Zhipu) · maker's own API$0.15$0.50$0.03$0.23751MTakes image and file input too. Cached-input storage is listed as limited-time free.Price page · Context
MiniMax-M3.1-FlashMiniMax · maker's own API$0.15$0.60$0.015$0.26251MCached figure is MiniMax's prompt-caching read price.Price page · Context
DeepSeek V4.1 FlashDeepSeek · maker's own API$0.30$1.20$0.006$0.5251MPeak rate. Off-peak (half price): $0.15 input, $0.60 output, $0.003 cached.Price page
MiniMax-M3 (up to 512K input)MiniMax · maker's own API$0.30$1.20$0.06$0.5251MShown at MiniMax's current "permanent 50% off" price; the list price is $0.60 input, $2.40 output. Above 512K input tokens the price is $0.60 input, $2.40 output, $0.12 cached, also after the same 50% off (list $1.20 / $4.80 / $0.24).Price page · Context
GLM-5.3-FlashXZ.ai (Zhipu) · maker's own API$0.37$1.25$0.075$0.591MThe faster Flash variant. Z.ai quotes about 200 tokens a second for it.Price page · Context
DeepSeek V4 ProDeepSeek · maker's own API$1.32$3.96$0.044$1.981MPeak rate. Off-peak (half price): $0.66 input, $1.98 output, $0.022 cached. No vision support.Price page
GLM-5.3Z.ai (Zhipu) · maker's own API$1.40$4.40$0.26$2.151MText input only, per Z.ai. Cached-input storage is listed as limited-time free.Price page · Context
Kimi K3.1 ProMoonshot (Kimi) · maker's own API$2.00$10.00$0.20$4.001,048,576Batch API is billed at 60% of the standard price. Cache writes are billed separately ($2.00 per 1M for 5 minutes, $4.00 for 1 hour).Price page · Context
Kimi K3Moonshot (Kimi) · maker's own API$3.00$15.00$0.30$6.001,048,576Moonshot calls K3 its previous-generation flagship. Batch API at 60% of standard. Cache writes: $3.00 per 1M for 5 minutes, $6.00 for 1 hour.Price page · Context
Kimi K3.1 Pro high-speedMoonshot (Kimi) · maker's own API$4.00$20.00$0.20$8.001,048,576The high-speed variant of K3.1 Pro. Cache writes: $4.00 per 1M for 5 minutes, $8.00 for 1 hour.Price page · Context

Scroll the table sideways on a phone. "Blended" is (3 × input + 1 × output) ÷ 4, the same mix as our cheapest AI models table. Prices are standard pay-as-you-go rates before tax.

Together AI reseller prices

Together AI hosts open-weight models and sets its own prices. These rows are reseller prices, not Moonshot's, Z.ai's, MiniMax's or DeepSeek's.

Reseller prices: what Together AI charges for the same open models, in USD per 1M tokens, checked 30 Oct 2026. These are not the makers' own prices.
ModelInput /1MOutput /1MCached input /1MBlended 3:1ContextAgainst the maker's priceSource · checked
GLM-5.3-FlashTogether AI · reseller price$0.15$0.50$0.03$0.23751MInput and output match the makerTogether AI page
MiniMax M3.1 FlashTogether AI · reseller price$0.15$0.60$0.015$0.26251MInput and output match the makerTogether AI page
DeepSeek V4.1 Flash 1017Together AI · reseller price$0.30$1.20$0.006$0.5251MInput and output match the maker (DeepSeek's peak rate)Together AI page
DeepSeek V4 Pro 0813Together AI · reseller price$1.32$3.96$0.13$1.981.05MInput and output match the maker (DeepSeek's peak rate); cached $0.13 vs $0.044 from the makerTogether AI page
GLM-5.3Together AI · reseller price$1.40$4.40$0.26$2.151MInput and output match the makerTogether AI page
Kimi K3Together AI · reseller price$3.00$15.00$0.30$6.001.05MInput and output match the makerTogether AI page

Which is cheapest for what?

This section only restates the sourced prices above, and compares price rather than quality.

What we couldn't verify or left out

We didn't find these on the official pages listed under Sources, or didn't check them, so we haven't guessed.

Sources

Every price on this page comes from the pages below, opened on .

FAQ

Which is cheapest: Kimi, MiniMax, GLM or DeepSeek?

By list price per token, GLM-5.3-Flash is lowest at $0.15 input and $0.50 output per 1M tokens, followed by MiniMax-M3.1-Flash at $0.15 and $0.60. DeepSeek V4.1 Flash is $0.30 and $1.20 at peak and $0.15 and $0.60 off-peak. Kimi's cheapest row, K3.1 Pro, is $2.00 and $10.00. This compares price only, not quality. [Z.ai docs: Pricing · MiniMax platform: Pricing Overview · DeepSeek API docs: Models & Pricing · Moonshot Kimi API: Model Inference Pricing]

How much does Kimi cost compared with DeepSeek?

Kimi K3.1 Pro is $2.00 input and $10.00 output per 1M tokens, and K3 is $3.00 and $15.00. DeepSeek V4 Pro is $1.32 and $3.96 at peak, and DeepSeek V4.1 Flash is $0.30 and $1.20. At a 3:1 input-to-output mix, K3.1 Pro blends to $4.00 per 1M tokens against $1.98 for DeepSeek V4 Pro at peak. [Moonshot Kimi API: Model Inference Pricing · DeepSeek API docs: Models & Pricing]

When is DeepSeek's off-peak discount?

DeepSeek treats 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays, as peak. Everything else is off-peak, at half the peak price: $0.15 input and $0.60 output for V4.1 Flash, and $0.66 and $1.98 for V4 Pro. [DeepSeek API docs: Models & Pricing]

Are Together AI prices the same as the makers' own prices?

For all six models we compared, Together's input and output prices match the maker's own. Together's cached price for DeepSeek V4 Pro is $0.13 per 1M tokens, against $0.044 from DeepSeek. Together is a reseller, so we label its rows as reseller prices and they can change independently. [Together AI: Kimi K3 (reseller) · Together AI: GLM-5.3 (reseller) · Together AI: GLM-5.3-Flash (reseller) · Together AI: MiniMax M3.1 Flash (reseller) · Together AI: DeepSeek V4.1 Flash 1017 (reseller) · Together AI: DeepSeek V4 Pro 0813 (reseller) · DeepSeek API docs: Models & Pricing]

What context window do these models have?

Each model in the tables lists a context window of about 1M tokens. Moonshot's is 1,048,576 tokens for the K3 series. MiniMax-M3 charges double above 512K input tokens, and Moonshot says Kimi K3.1 Pro and K3 have flat pricing across the window. [Moonshot Kimi API: Model Inference Pricing · MiniMax platform: Pricing Overview · MiniMax platform: Models · Z.ai docs: GLM-5.3 · Z.ai docs: GLM-5.3-Flash/FlashX · DeepSeek API docs: Models & Pricing]

How does cached-input pricing compare?

Cache-hit input per 1M tokens: DeepSeek V4.1 Flash $0.006, MiniMax-M3.1-Flash $0.015, GLM-5.3-Flash $0.03, DeepSeek V4 Pro $0.044, Kimi K3.1 Pro $0.20, GLM-5.3 $0.26 and Kimi K3 $0.30. Moonshot also charges for cache writes on its K3 models, which the other pricing pages we read don't list. [DeepSeek API docs: Models & Pricing · MiniMax platform: Pricing Overview · Z.ai docs: Pricing · Moonshot Kimi API: Model Inference Pricing]

What does this page leave out?

Subscription products, older Kimi K2 models, other currencies and regions, Together AI rows for Kimi K3.1 Pro and MiniMax-M3, rate limits, and any quality or cost-per-task comparison. We left them out because we didn't verify them, and the section above lists each one. [Moonshot Kimi API: Model Inference Pricing · Z.ai docs: Pricing · MiniMax platform: Pricing Overview · DeepSeek API docs: Models & Pricing]