Model comparisons
Beam 1.0 bills 77% less than Kimi K3 per million output tokens. Tier prices, request fields and limits side by side.
Kimi K3 is Moonshot AI's long-context agent model behind an OpenAI-compatible chat call. It reads 1,000,000 tokens, replies up to 1,048,576, with reasoning effort low, high or max, up to 128 tools and image input.
View model details →View sample outputs →Beam 1.0 from Reflection supports coding and long-running tool workflows. This chat API accepts text, provides a 1,048,576-token input budget and accepts max_tokens up to 943,718. Stream Beam 1.0’s text as it arrives, or let a function-call result supply the evidence for the next step in long coding sessions with Beam 1.0.
View model details →View sample outputs →Compare
Start with the input you have, the result you want and your budget. These are the differences that affect that choice.
Compare the published examples, then try the same input on both models with matching settings.
The Spicy Index and Freedom Score for both models, from the same daily calculation as the leaderboards.
| Measure | Kimi K3 | Beam 1.0 |
|---|---|---|
| Spicy Index | 77.8 | 68.9 |
| Freedom Score | 74.4 | Not yet tested |
| Evidence | Tested 5 of 5 probe types · 24 runs | — |
| Range if the rest pass | 74.4–74.4 | — |
| Price | $5.40 per 1M mixed tokens (3 input : 1 output) | $1.50 per 1M mixed tokens (3 input : 1 output) |
| Reviews | Read the Kimi K3 review | Read the Beam 1.0 review |
Preliminary · freedom evidence pending
See both on the leaderboard →Current price
A start price answers a different question than the one you have. These are the tiers both models publish, matched setting for setting, in US dollars per billed unit with no conversion step in between.
Swipe horizontally to compare all fields.
| Tier | Kimi K3 | Beam 1.0 | Difference |
|---|---|---|---|
| Input | $2.7 / 1 million tokens | $1 / 1 million tokens | Beam 1.0 −62% |
| Cached input | $0.3 / 1 million tokens | $0.1 / 1 million tokens | Beam 1.0 −66% |
| Output | $13.5 / 1 million tokens | $3 / 1 million tokens | Beam 1.0 −77% |
Arithmetic on the published rates at each model's own default setting, not a quote. The console prices your actual settings before a request is accepted.
About this comparison
Every row here is read from the JSON Schema the endpoint validates against, so it matches what the API will really accept rather than a feature list written by hand.
Swipe horizontally to compare all fields.
| Controls and limits | Kimi K3 | Beam 1.0 |
|---|---|---|
| Supported tasks | chat | chat |
| Input context | 1,000,000 tokens | 1,048,576 tokens |
| Maximum output | 1,048,576 tokens | 943,718 tokens |
| Tool calling | Yes | Yes |
| Execution | Synchronous, with streaming | Synchronous, with streaming |
| Request timeout | 60 minutes | 10 minutes |
| Availability | Live | Live |
Do low-filter models allow every request?
Two separate facts sit behind these labels. The first is what we do, which is nothing: no platform content filter, and no maturity flag to set on a request. The second is what each model does on its own, and that is the only part that varies between these two.
Kimi K3
We add no content filter to this model, but like most chat models it declines a blunt request that comes with no context. It writes unrestricted fiction reliably once your system prompt sets up a fiction-writing context, or once the request is framed as a story. Legal and consent duties remain with the customer.
Usage note For Kimi K3, send only material you are authorised to process, respect privacy and identity, and obtain consent; adult-content requests must involve adults only.
Use these content labels to shortlist models, then open each model page for available examples and content-test details.
The label never changes your obligations. Age, consent, real-person likeness and the law where you operate stay with you, and SpicyAPI does not inspect each request.
Can I swap the model ID without changing the request body?
The task lifecycle is shared; the body is not. Each endpoint publishes its own JSON Schema and the API validates against that, so this is the list to reconcile before a switch.
How to choose
Kimi K3 can return more in one reply: up to 1,048,576 tokens against the other model's ceiling.
Beam 1.0 takes the longer prompt: up to 1,048,576 input tokens in a single request.
Compare
The model identifiers are moonshot/kimi-k3/chat and reflection/beam-1.0/chat, and both go in the same field of the same request.
Both return the response on the chat request. With streaming enabled, read the stream until it finishes; there is no media task to poll.
Both use the same submit-and-track flow, but their required inputs may differ. Check each model page before switching.
One key, one balance, one set of spend caps. Both bill to the same account in US dollars, and the per-key daily, monthly and lifetime limits apply to whichever one you call.
Questions developers ask
Beam 1.0 costs 77% less per million output tokens. Compare the input and cache rates below for longer conversations. Prices may change; check the estimate shown for your actual settings before submitting.
Kimi K3 allows 60 minutes per task and Beam 1.0 allows 10. Those are ceilings rather than typical times — most requests finish far sooner — but they are what your own client timeout has to survive.
Prices and capabilities shown here come from the current public model catalogue. Check each model page for its latest inputs and examples.