> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kymaapi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# FAQ

> Frequently asked questions about Kyma API.

## Pricing & access

<AccordionGroup>
  <Accordion title="Is Kyma really free?">
    Yes to start. You get $0.50 free credits on signup. No credit card required. That's enough to try Kyma on lower-cost models. For heavier usage, buy credits starting at $5.
  </Accordion>

  <Accordion title="What happens when I hit the rate limit?">
    You'll get a `429` response. Wait a few seconds and retry. Limits are tier-based: Tier 0 = 30 RPM, up to Tier 4 = 300 RPM. See [Rate Limits](/pricing#rate-limits).
  </Accordion>

  <Accordion title="Can I increase my rate limit?">
    Yes. Rate limits increase automatically as you purchase credits. Tier 1 unlocks at \$10 purchased (60 RPM). See [Rate Limits](/pricing#rate-limits) for all tiers.
  </Accordion>
</AccordionGroup>

## Models & capabilities

<AccordionGroup>
  <Accordion title="Which model should I use?">
    Start with `qwen-3.6-plus` if you want the safest default. Use `kimi-k2.6` for tool-heavy agents, `deepseek-r1` for hard reasoning, and `gemini-2.5-flash` for long-context tasks. See [Which model should I use?](/models/recommended) for the full decision guide.
  </Accordion>

  <Accordion title="Are these models truly open source?">
    Yes. All models served by Kyma are open-source or open-weight models from Meta, Alibaba, Google, OpenAI, and others. Kyma itself is a managed service — not open source.
  </Accordion>

  <Accordion title="Do you support vision or multimodal inputs?">
    Yes. `gemma-4-31b` is the cheapest strong multimodal option, and `kimi-k2.6` is the better pick if you also need tool-heavy agent behavior.
  </Accordion>

  <Accordion title="Can I generate images?">
    Yes. Kyma serves four image-generation models — `flux-1.1-ultra` (cinematic photo), `flux-kontext-pro` (image edit), `ideogram-v3` (typography / logos), and `recraft-v3` (vector / illustration) — through async [`POST /v1/images/generations`](/api-reference/images-generations). Pricing is per-image, $0.054 to $0.108 depending on model. See the [Image Generation guide](/models/image-generation) for prompts and examples.
  </Accordion>

  <Accordion title="What is the maximum context window?">
    Up to 1M tokens with Gemini models. Many other active models support 128K-262K. Check [`GET /v1/models`](/api-reference/models-list) for the live catalog and context windows.
  </Accordion>
</AccordionGroup>

## Technical limits

<AccordionGroup>
  <Accordion title="Does Kyma have a hidden output token limit?">
    No. Kyma does not impose any output token cap on the gateway. The `max_tokens` parameter from your request is forwarded directly to the model. The `max_output_tokens` value shown in `/v1/models` reflects the model creator's published specification, not a Kyma restriction. You can verify this via `GET /v1/capabilities`.
  </Accordion>

  <Accordion title="Are models on Kyma different from the original?">
    No. Kyma serves the exact same model weights published by the creator (Alibaba, Google, DeepSeek, Meta, etc.). The models run on high-performance inference infrastructure. Kyma does not fine-tune, quantize, or modify any model.
  </Accordion>

  <Accordion title="Why did my response get cut short?">
    Three common causes:

    1. **You didn't set `max_tokens`** — defaults to 4096 tokens. Set it higher for longer outputs (e.g., `max_tokens: 8192`).
    2. **Model's output limit** — each model has a maximum output capacity (check `max_output_tokens` in `/v1/models`). For long generation, use models like `glm-5.1` (65K) or `deepseek-r1` (32K).
    3. **`finish_reason: length`** — the model hit its limit naturally. Increase `max_tokens` or use a model with higher output capacity.

    Kyma never truncates output. If `finish_reason` is `length`, the model itself stopped.
  </Accordion>

  <Accordion title="What is the default max_tokens if I don't set it?">
    4096 tokens. Always set `max_tokens` explicitly in your request if you need longer outputs. Kyma forwards your value directly to the model.
  </Accordion>

  <Accordion title="Where can I see all model limits?">
    Call `GET /v1/models` — each model includes `max_output_tokens`, `context_window`, `gateway_output_limit` (always `null`), and `max_tokens_passthrough` (always `true`). For a gateway-level overview, call `GET /v1/capabilities`.
  </Accordion>
</AccordionGroup>

## Compatibility

<AccordionGroup>
  <Accordion title="Is Kyma compatible with the OpenAI SDK?">
    Yes. Just change `base_url` to `https://kymaapi.com/v1` and use your `ky-` API key. All OpenAI SDK features work.
  </Accordion>

  <Accordion title="Does Kyma work with the Anthropic SDK?">
    Yes. Kyma supports the Anthropic Messages API at `/v1/messages`. See our [Anthropic guide](/guides/anthropic).
  </Accordion>

  <Accordion title="Can I use Kyma with LangChain?">
    Yes. Use the OpenAI provider in LangChain with Kyma's base URL.
  </Accordion>

  <Accordion title="Can I use Kyma in Cursor, Claude Code, or Continue.dev?">
    Yes. Configure Kyma as an OpenAI-compatible provider with your `ky-` API key and `https://kymaapi.com/v1` as the base URL.
  </Accordion>

  <Accordion title="Can I use Kyma with Google Antigravity?">
    The official Google Antigravity IDE does not support custom API endpoints — it only works with Gemini, Claude Sonnet, and GPT-OSS via Google's auth. However, you can use the community [open-antigravity](https://github.com/ishandutta2007/open-antigravity) fork which supports custom OpenAI-compatible endpoints like Kyma. See our [Antigravity guide](/guides/antigravity) for setup instructions.
  </Accordion>
</AccordionGroup>

## Reliability

<AccordionGroup>
  <Accordion title="What happens when a model's provider is down?">
    Kyma automatically retries your request on backup infrastructure. Most outages are invisible to you — your request succeeds with the same model from a different source. If the exact model is unavailable everywhere, Kyma can substitute an equivalent-quality model. Check the `X-Kyma-Fallback` response header to know if a fallback was used.
  </Accordion>

  <Accordion title="What are the X-Kyma-Fallback headers?">
    When a fallback occurs, these headers are included in the response:

    * `X-Kyma-Fallback: true` — a fallback was used
    * `X-Kyma-Fallback-Layer: 1|2|3` — 1 = same model different source, 3 = different model

    Use `X-Kyma-Fallback-Layer` to decide whether to retry. Layer 1 means you got the same model you asked for; Layer 3 means an equivalent-quality substitute.
  </Accordion>

  <Accordion title="What is Kyma's uptime?">
    Kyma runs on redundant infrastructure with auto-failover. Individual provider outages don't affect you because requests are automatically rerouted. Check [Status](https://kymaapi.com/status) for real-time model availability.
  </Accordion>
</AccordionGroup>

## API keys & account

<AccordionGroup>
  <Accordion title="How do API keys work?">
    Keys start with `ky-`. You can create multiple keys in the [Dashboard](https://kymaapi.com/dashboard/keys). Each key shares the same account balance and rate limits.
  </Accordion>

  <Accordion title="Is my data private?">
    Kyma does not store your prompts or responses. We only log metadata (model, latency, tokens) for usage tracking.
  </Accordion>

  <Accordion title="Can I use this in production?">
    Yes. Higher tiers unlock automatically as you purchase credits. Tier 1+ gets 60–300 RPM, credits never expire, and you can switch models without changing your integration.
  </Accordion>
</AccordionGroup>