macrostack

AI API price index · the same model at every host · read 10 Nov 2026 to 17 Nov 2026

Kimi K3 API pricing at every host

8 hosts sell Kimi K3 at a published per-token price. Blended, it costs $5.70 to $6.00 per million tokens — 1.1× apart for the same model. Cheapest: DeepInfra. Each price read on the host’s own pricing page and linked to it.

How this is ranked

The same model, so a cheapest is honest. Hosts are ordered by blended price, (3 × input + 1 × output) ÷ 4 per million tokens, then by output, then input, then name. Equal prices are shown as equal, never as one host winning.

The index decides what counts. Standard tier, global rate, list price, standard prompt length — the same rules as the AI API price index. Every other price for Kimi K3 stays on this page and says why it is not ranked. A page needs 3 ranked hosts to exist at all.

Kimi K3 at 8 hosts, cheapest blended first

Per 1M tokens, read 10 Nov 2026 to 17 Nov 2026.

HostInputCached inputOutputBlendedWhat the host’s page records
DeepInfraKimi-K3$2.85$0.285$14.25$5.70context 1024kread 17 Nov 2026
Amazon BedrockKimi K3$3.00$0.30$15.00$6.00On-demand, Global cross-Region inferenceread 14 Nov 2026
Fireworks AIKimi K3$3.00$0.30$15.00$6.00Tier: serverless · Standard mode; Priority $3.75 / $0.375 / $18.75read 10 Nov 2026
Moonshot AI (Kimi)kimi-k3$3.00$0.30$15.00$6.00K3 series, 1,048,576-token context; cache writes $3.00 (5m) and $6.00 (1h) per 1M tokensread 17 Nov 2026
Nebius Token FactoryKimi-K3$3.00$0.30$15.00$6.00public endpoint (shared; the page says suitable for testing, not production)read 17 Nov 2026
Novita AIKimi K3$3.00$0.30$15.00$6.00Tier: serverless · Context 1M; cache write (5m) $3, cache write (1h) $6read 17 Nov 2026
Perplexityperplexity/kimi-k3$3.00$0.30$15.00$6.00open-weight model hosted by Perplexityread 17 Nov 2026
Together AIKimi K3$3.00$0.30$15.00$6.00Tier: serverless · List price. Re-read 16 Oct: the promotion the 10 Oct page printed ('Promotion valid until 10/11/2026', $2.70 / $0.27 cached / $13.50) has ended; the page and the docs model table both print $3.00 / $0.30 cached / $15.00.read 10 Nov 2026

Blended = (3 × input + 1 × output) ÷ 4. A dash means the host prints no cached-input price for this model.

Where the hosts differ

Only what each host's own price row records. Nothing here is inferred.

  • Context length: DeepInfra, context 1024k; Moonshot AI (Kimi), 1,048,576-token context; Novita AI, Context 1M. The other 5 hosts' rows do not record it.
  • Speed: no host's row records it.
  • Quantisation: no host's price row records it — ask before assuming two endpoints run identical weights.
  • Use: Nebius Token Factory, "suitable for testing, not production".

Also priced, not ranked

Other prices the index holds for Kimi K3, each with the reason it is left out of the ranking.

HostInputOutputWhy not ranked
Amazon BedrockOn-demand, US cross-Region inference$3.30$16.50in-Region price (10% above the global rate); the index ranks the global price

Read next

Questions

How much does the Kimi K3 API cost?
Read on each host's own page (10 Nov 2026 to 17 Nov 2026), Kimi K3 costs from $5.70 to $6.00 per million tokens blended at 8 hosts — 1.1× apart for the same model. The cheapest is DeepInfra at $2.85 in and $14.25 out per million tokens; the dearest is Together AI at $3.00 in and $15.00 out per million tokens.
Which host is cheapest for Kimi K3?
DeepInfra, at $5.70 per million tokens blended ($2.85 in and $14.25 out per million tokens). The other 7 hosts charge more for the same model.
Is Kimi K3 the same model at every host?
The model is the same open-weight release, matched by name on each host's price list. What the hosts' own price rows record differs: context length (DeepInfra: context 1024k; Moonshot AI (Kimi): 1,048,576-token context; Novita AI: Context 1M), use (Nebius Token Factory: "suitable for testing, not production"). None of these price rows records quantisation, so ask the host before assuming two endpoints run identical weights.
What does blended price mean here?
Blended = (3 × input + 1 × output) ÷ 4, per million tokens: three input tokens to every output token, a common mix for chat and retrieval. It ranks hosts on one number; the input, cached-input and output prices are all on this page for any other mix.
Can I use Kimi K3 free?
No host in the index names Kimi K3 among its free models or free-tier models. Fireworks AI give a one-time starting credit that could be spent on it; a credit runs out.

Prices read 10 Nov 2026 to 17 Nov 2026 on each host’s own pricing page. Macrostack measures nothing and is paid by no host on this page. Part of the AI API price index on Layer 2.5 — Compute.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.