Inference providers · open weights · Moonshot AI
Kimi K3
Open-weights model published by Moonshot AI, which also serves it through its own endpoint on OpenRouter: that rate is the maker's list price, and the other 5 hosts below are compared with it. Each figure was read on that host's own pricing page and links to it, with the date it was checked.
What each host charges
6 hosts, from $2.85 (DeepInfra) to $3.00 (Baseten, Fireworks AI, Novita AI, OpenRouter and Together AI) per 1M input tokens: a 1.1× spread, within the 1.15× that AI Signal treats as the same price. Pick a host on latency, region or tooling; the bill will be the same.
| Host | Input | Output | Cached input | Context served | Checked | Source |
|---|---|---|---|---|---|---|
| DeepInfra | $2.85 | $14.25 | $0.285 | 1.0M tokens | November 16, 2026 | deepinfra.com |
| Together AI | $3.00 | $15.00 | $0.30 | 1.0M tokens | November 16, 2026 | together.ai |
| Fireworks AI | $3.00 | $15.00 | $0.30 | 1.0M tokens | November 16, 2026 | docs.fireworks.ai |
| OpenRouter (via Moonshot AI, the maker) | $3.00 | $15.00 | $0.30 | 1.0M tokens | November 18, 2026 | openrouter.ai |
| Novita AI | $3.00 | $15.00 | $0.30 | 1.0M tokens | November 16, 2026 | novita.ai |
| Baseten | $3.00 | $15.00 | $0.30 | — | November 16, 2026 | baseten.co |
DeepInfra charges $2.85 per 1M input tokens and Baseten, Fireworks AI, Novita AI, OpenRouter and Together AI $3.00 for the same weights; on output, $14.25 against $15.00.
Mentioned in
4 posts on AI Signal cite Kimi K3. The sentence is the post's own, taken from where it links here.
-
AI Pricing This Week: Large 4 Cut, New Hosts, Oct 26-Nov 1 · November 7, 2026
Together raised the cached-input price of Kimi K3 from $0.27 to $0.30 per million tokens; input ($3.00) and output ($15.00) did not move.
-
Together, Fireworks, DeepInfra, Groq and Novita: What Open-Model Hosts Say Besides the Token Price · October 28, 2026
They are on open-model prices by host, with one page per model for those that pass AI Signal's quality filter, such as GLM-5.3, gpt-oss-120b and Kimi K3.
-
AI Pricing This Week: Gemini 4 Argon, Muse Wonder, Oct 12-18 · October 19, 2026
The Kimi K3 page shows the corrected figures.
-
Same Open Model, Different Host: What a Month Costs · October 12, 2026
| Model | Hosts priced | Cheapest host, month | Most expensive host, month | Difference | |---|---|---|---|---| | Llama 3.3 70B Instruct | 2 | DeepInfra $16.40 | Together AI $124.80 | $108.40 (7.61×) | | DeepSeek V4 Flash 0731 | 2 | DeepInfra $9.60 | Together AI $19.60 | $10.00 (2.04×) | | Kimi K3 | 4 | DeepInfra $570.00 | Together AI, Fireworks AI, OpenRouter $600.00 | $30.00 (1.05×) | | GLM-5.3 | 3 | $228.00 on every host | $228.00 on every host | none | | gpt-oss-120b | 3 | $27.00 on every host | $27.00 on every host | none | | MiniMax M3 | 3 | $54.00 on every host | $54.00 on every host | none |
Recorded price changes
Every change AI Signal has recorded for a host's price of Kimi K3, with the date it took or takes effect and why. Changes of our own rule are listed too, and say so.
-
October 29, 2026 · DeepInfra · correction of our own reading
Correction of our own reading, not a price change by DeepInfra. DeepInfra's pricing page prints a cached input price inside the input cell of some models ('$0.06 / $0.015 cached'), and two offerings of this dataset did not carry it: DeepSeek V4 Flash 0731 ($0.015) and Kimi K3 ($0.285) per million tokens, read on October 29, 2026. Both rows already printed it when read on September 30, 2026, as the quotes in their notes show; the figure was not carried into the offering. Input and output prices are unchanged.
-
October 17, 2026 · Together AI · price changed, see below
Together AI raised the cached input price of Kimi K3 from $0.27 to $0.30 per million tokens; input ($3.00) and output ($15.00) are unchanged, so the headline price does not move. The last reading at $0.27 was AI Signal's hand reading of October 12, 2026, 'Kimi K3 | $3.00 $0.27 (cached) | $15.00 |'; AI Signal's next saved reading of the page, on October 17, 2026, already printed 'Kimi K3 | $3.00 $0.30 (cached) | $15.00 |', as did the weekly detector on October 19 and 26 and the page on October 29, 2026. The exact day of the change is not known, so the date here is the first day it was seen. Correction, October 30, 2026: this entry was dated October 19 until then; the October 17 reading had been missed.
-
October 12, 2026 · Together AI · correction of our own reading
Correction of our own reading, not a price change by Together AI. From October 6 to October 12, 2026 AI Signal published Kimi K3 on Together AI at $2.70 input and $13.50 output per million tokens and recorded it as a 10% price cut by Together. It was a promotion: on October 12, 2026 Together's pricing page labels the row 'PROMO' with 'Promotion valid until October 11, 2026' and prints the list price beside it, $3.00 input and $15.00 output, the price it had until October 5. The list price did not change and AI Signal publishes list prices, so the row is back at $3.00 / $15.00. Cached input stays at $0.27, the figure Together's Standard table prints on October 12, 2026 (it was $0.30 until October 5). This entry replaces the one of October 6 that called the promotion a price cut.
-
September 1, 2026 · Fireworks AI (all models on this host) · price up 1.5×
Since September 1, 2026 Fireworks adds a 50% surcharge to the base price of the Serverless models it marks as US-only. The Fireworks prices on AI Signal are the base prices, without the surcharge.
On OpenRouter
On September 15, 2026 OpenRouter listed 20 endpoints for Kimi K3, 18 of them healthy, from $2.10 (InferenceNet) to $6.00 per 1M input tokens. The figure OpenRouter's catalog shows for the model, $2.6481 / $13.2827 (Sail Research), is the 6th cheapest of those 20. Of these, only Moonshot AI's own endpoint is the maker's list price; the rest are resellers. DeepInfra, Together AI, Fireworks AI and Baseten, priced above from their own pages, are among them.
Why the catalog figure is not what a request pays →
From Moonshot AI's model card
- Size: “It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window.” (model card on Hugging Face, read October 11, 2026)
- Reasoning effort: “Kimi K3 always has thinking enabled, and will return reasoning_content. Thinking effort is configured with the top-level reasoning_effort request field, which supports "low", "high", and "max" (default "max" ).” (model card on Hugging Face, read October 11, 2026)
- License: “Both the code repository and the model weights are released under the Kimi K3 License.” (model card on Hugging Face, read October 11, 2026)
Specifications
| Weights published by | Moonshot AI |
|---|---|
| Context length (maker) | 1.0M tokens |
| Context served by hosts | 1.0M tokens |
| API model ids | DeepInfra: moonshotai/Kimi-K3; Together AI: moonshotai/Kimi-K3; Fireworks AI: kimi-k3; OpenRouter: moonshotai/kimi-k3; Novita AI: moonshotai/kimi-k3; Baseten: moonshotai/Kimi-K3 |
| Hosts with a verified price | 6 |
Host prices verified between November 16, 2026 and November 18, 2026, each on the host's own pricing page linked in the table above, never from a reseller catalog or a third-party table. Maker and context length from Moonshot AI's model card, checked September 15, 2026 (max_position_embeddings 1048576 in the model's config.json (text_config)). How the data is checked.
Every open model, every host, one table →
When comparing hosts is worth it, and when it is not →
What the OpenRouter price means, and what a request actually pays →