--- title: Hero AI Models slug: turbine-user-guide/hero-ai-models docTags: createdAt: 2026-05-20T12:00:00.000Z --- ## Overview Hero AI does not use a single model for every feature. For default **Swimlane-managed** routing, inference uses **Anthropic Claude** on **Amazon Bedrock**, with a different primary model per feature. **Hero AI Native Action** defaults to **Claude Haiku 4.5**. Starting in **Turbine 26.4.0**, **Agents** (plan generation, remediation, playbook builder, and component agents) default to **Claude Sonnet 5**, with **Claude Sonnet 4.5** as fallback. **Hero AI Companion** stays on **Claude Sonnet 4.5**. Starting in **26.2.0**, you choose the model for each [Hero AI Native Action](docId\:C4CKmT7fHXgjBtlXxOOC2) on the **Prompt** tab (**Model Selection**), or leave **Use Default Model** for **Claude Haiku 4.5**. If a saved **modelId** is no longer in the available models list (for example after a [Custom LLM](docId\:Deoy-K3jmcpdm-c-IKRQ4) provider change), the action uses the **Small** model at run time and shows **Previously Selected Model is not available** in the action dialog. See **When the Saved Model Is Unavailable** in [Hero AI Native Action](docId\:C4CKmT7fHXgjBtlXxOOC2). You can replace Swimlane-managed models with [Custom LLM](docId\:Deoy-K3jmcpdm-c-IKRQ4) or **Turbine Platform** installer settings. **Swimlane does not charge Hero AI Credits for usage through your Custom LLM models.** Use the sections below for default models by feature, **data residency** when you pick models, and how **tokens convert to credits** on Swimlane-managed models. ## Default Swimlane Models For standard **Turbine** deployments, Swimlane routes these features to the primary Claude model on Bedrock when you use Swimlane-managed routing: | Feature | Default (primary) | | ----------------------------------------------------------------------------- | -------------------------------------------- | | [Hero AI Native Action](docId\:C4CKmT7fHXgjBtlXxOOC2) | Claude Haiku 4.5 | | [Hero AI Companion](docId\:crQKFVnGpCz_wJ8xtHDs7) | Claude Sonnet 4.5 | | **Agents** (plan generation, remediation, playbook builder, component agents) | Claude Sonnet 5 (fallback Claude Sonnet 4.5) | :::hint{type="info"} **Use Default Model** on a [Hero AI Native Action](docId\:C4CKmT7fHXgjBtlXxOOC2) is **Claude Haiku 4.5**. **Claude Sonnet 5** is the default for **Agents**. **Claude Sonnet 4.5** is the default for [Hero AI Companion](docId\:crQKFVnGpCz_wJ8xtHDs7). Sonnet 5 is not offered in the Native Action **Model** drop-down in Turbine 26.4. ::: **Hero AI Companion** uses **Claude Sonnet 4.5** on Swimlane-managed routing, as shown in the table above. **Agents** use **Claude Sonnet 5** for AI SOC plan and remediation flows, Playbook Generation, and related agentic workflows. If **Claude Sonnet 5** is unavailable, Swimlane-managed routing falls back to **Claude Sonnet 4.5**. **Hero AI Companion** continues to use **Claude Sonnet 4.5** and does not support Sonnet 5. If Custom LLM is enabled and the Medium default is Sonnet 5, Companion does not work. Keep Medium on Sonnet 4.5 when users need Companion. You cannot configure Agent or Companion routing in the UI. Swimlane does not use your content to train or fine-tune these foundation models. ## When Defaults Do Not Apply The table above describes **Swimlane-managed default routing**. Your environment may use different models when: | Override | Applies to | Details | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | [Custom LLM](docId\:Deoy-K3jmcpdm-c-IKRQ4) | **Turbine Cloud** (LiteLLM) and **Turbine Platform** (LiteLLM or Custom Bedrock) | Account-level **Custom LLM** settings replace default Bedrock routing with your proxy or Bedrock inference profiles. Usage through Custom LLM is not billed as Hero AI Credits. | | [Turbine Platform Installer](https://docs.swimlane.com/turbine-installer/configure-the-turbine-platform-for-an-embedded-cluster-install#llm-proxy-settings) (**LLM Proxy Settings**; Turbine Platform **26.0.4** only) | **Turbine Platform** on-premises | Environment variables (for example `LLM_MODEL_ID_MEDIUM`, `SWIMLANE_HEROAICHAT__Models__BedRock__Model`) can override defaults. Confirm with your Swimlane contact which settings apply in your cluster. | | **Hero AI Native Action model selection** | Playbooks that set an explicit model on the **Prompt** tab | Starting in **26.2.0**, choose a model in **Model Selection** or leave **Use Default Model** (**Claude Haiku 4.5**). The selected model overrides Swimlane-managed default routing for that action only. See [Hero AI Native Action model residency](https://docs.swimlane.com/hero-ai-models#hero-ai-native-action-model-residency-2620). | ## Regional Availability Residency labels in the tables below describe where inference runs for your Swimlane instance: | Label | Meaning | | ------------------------------ | --------------------------------------------------------------------------------------------- | | **US**, **EU**, **AU**, **JP** | Inference stays within that geography for your instance. | | **Global** | Inference may route across AWS Regions when residency rules allow. | | **In-region** | The model runs in the AWS Region for that Swimlane instance (not a cross-region geo profile). | | **N/A** | The model is not offered for that instance. | ### Default Models (Haiku 4.5 and Sonnet 4.5) The table below shows where **Claude Haiku 4.5** (default when **Hero AI Native Action** uses **Use Default Model**) and **Claude Sonnet 4.5** (default for [Hero AI Companion](docId\:crQKFVnGpCz_wJ8xtHDs7); Agents fallback in 26.4.0) run for each Swimlane instance. | Swimlane instance | Claude Haiku 4.5 | Claude Sonnet 4.5 | | -------------------- | ---------------- | ----------------- | | **usn.swimlane.app** | US | US | | **us1.swimlane.app** | US | US | | **gov.swimlane.app** | US | US | | **uk1.swimlane.app** | EU | EU | | **de1.swimlane.app** | EU | EU | | **ca1.swimlane.app** | US | US | | **au1.swimlane.app** | AU | AU | | **sg1.swimlane.app** | Global | Global | | **jp1.swimlane.app** | JP | JP | For AWS Region details within each geo profile, see **Geo inference details** on the matching model card: - [Claude Haiku 4.5 on Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-haiku-4-5.html) - [Claude Sonnet 4.5 on Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-anthropic-claude-sonnet-4-5.html) ### Hero AI Native Action Model Residency (26.2.0) Starting in **26.2.0**, you can select additional models on the **Hero AI Native Action** **Prompt** tab. The table below shows residency for each optional model by Swimlane instance. See the [residency labels](https://docs.swimlane.com/hero-ai-models#regional-availability) table for **US**, **EU**, **Global**, and **In-region**. | Model | usn | us1 | gov | uk1 | de1 | ca1 | au1 | sg1 | jp1 | | ----------------------- | --------- | --------- | --------- | --------- | --------- | --- | --------- | ------ | --------- | | **Claude Haiku 4.5** | US | US | US | EU | EU | US | AU | Global | JP | | **Claude Sonnet 4.5** | US | US | US | EU | EU | US | AU | Global | JP | | **OpenAI gpt-oss-20b** | In-region | In-region | In-region | In-region | In-region | N/A | In-region | N/A | In-region | | **OpenAI gpt-oss-120b** | In-region | In-region | In-region | In-region | In-region | N/A | In-region | N/A | In-region | | **Gemma 3 27B** | In-region | In-region | In-region | In-region | In-region | N/A | In-region | N/A | In-region | | **Qwen3 32B** | In-region | In-region | In-region | In-region | In-region | N/A | In-region | N/A | In-region | | **Claude Sonnet 4.6** | US | US | US | In-region | EU | US | AU | Global | JP | | **Claude Opus 4.5** | US | US | US | EU | EU | US | Global | Global | Global | | **Claude Opus 4.6** | US | US | US | In-region | EU | US | AU | Global | Global | For open-weight models, use [Regional Availability in Amazon Bedrock](https://docs.aws.amazon.com/bedrock/latest/userguide/models-region-compatibility.html) to confirm which models are offered in each AWS Region. ## How Credits Are Calculated When **Hero AI Credits** is enabled for your account and inference uses **Swimlane-managed** models, Turbine converts token usage into credits. At a high level: 1. The platform records token counts by category: **input**, **output**, **cache read**, and **cache write** (when caching applies). 2. Each category is multiplied by a **weight** for that model. 3. Weighted tokens are divided by the model’s **Tokens per credit** rate to produce credit consumption. For Swimlane-managed **Claude** models in the catalog (Haiku, Sonnet, Opus), the seeded weights are: | Category | Weight | | ----------- | ------ | | Input | 1 | | Output | 5 | | Cache write | 1.3 | | Cache read | 0.1 | So output tokens count **five times** as much as input tokens toward credits for those Claude models. Open-weight models in the catalog use similar input weight **1**, with output weights of **4.3** (gpt-oss-20b) or **4** (gpt-oss-120b, Qwen3 32B), and **0** for cache categories when caching does not apply. You do not configure these weights in the product UI. :::hint{type="info"} On the **Hero AI Native Action** **Model** drop-down, models can show a secondary line such as `Claude / 300 Tokens per Credit` so you can compare relative cost when selecting a model. See [Hero AI Native Action](docId\:C4CKmT7fHXgjBtlXxOOC2). **Sonnet 5** is filtered out of that drop-down in Turbine 26.4. ::: ## Usage (Tokens and Credits) Hero AI consumption is reported under **Admin Panel > Usage**: | Report | What it shows | | ----------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | | [Hero AI Credits](docId\:FPBH2J8Xm1TfDvEjnc8KB) | **Credit** usage and contract caps for the current billing period — **Playbook Usage** and **User Prompts** tabs | | [Hero AI Prompts](docId\:RV7WVN3wx2Uwa_bCHY7uR) | **Prompt counts** by playbook, component, or user | | [Hero AI Tokens](docId:8g7pp4ZUIwUisuVyQ32s9) | **Token** volume (input, output, cache read, cache write) | ### Model and Tokens Per Credit When your account uses **Hero AI Credits**, Turbine converts token usage to credits using the rates below. **Tokens per credit** is how many tokens of that model equal one credit before category weights are applied. Output tokens count more heavily than input tokens when credits are calculated. Cache read and cache write token counts appear on the [Hero AI Tokens](docId:8g7pp4ZUIwUisuVyQ32s9) dashboard; you do not configure those weights in the product. | Model | Tokens per credit | | ----------------------- | ----------------- | | **Haiku 4.5** | 300 | | **Opus 4.5** | 60 | | **Opus 4.6** | 60 | | **OpenAI gpt-oss-20b** | 4500 | | **OpenAI gpt-oss-120b** | 2200 | | **Qwen3 32B** | 2200 | | **Sonnet 4.5** | 100 | | **Sonnet 4.6** | 100 | | **Sonnet 5** | 100 | These rates apply to **Swimlane-managed** models on **Turbine Cloud** when **Hero AI Credits** is enabled for your account. **Swimlane does not charge Hero AI Credits for usage through **[Custom LLM](docId\:Deoy-K3jmcpdm-c-IKRQ4)**.** Credits, prompts, and tokens measure different aspects of Hero AI usage. Use the dashboard that matches the metric you need.