LLM models & AI Credits
Choose and save a model, understand the default, and compare quality, speed, and AI Credit usage. English narration and captions; pricing examples from 2 October 2026.
Before you start
Open an agent you own. The framework, language model, and provider route are separate choices.
Written guide
Choose a suitable model
Review the available models and their descriptions. Start with a model suited to the task, then compare results on the same small request.
Save the selection
Save the model in the agent configuration. Reopen the configuration to confirm the chosen value persisted.
Compare quality and usage
Review the output alongside recorded usage. Longer prompts, larger responses, and repeated tool calls can change the total cost of a task.
Check current prices
Use the current model catalog and billing information for purchasing decisions. The video shows pricing examples recorded on 2 October 2026.
A model change does not install skills or grant permissions. With BYOK, provider charges and Hatcher plan entitlements remain separate.
Video transcript
Your agent's model affects how it reasons how quickly it responds and how many AI Credits it uses In this guide we'll choose a model in Hatcher understand the default and compare the cost of everyday work with more demanding tasks First separate three things The framework such as OpenClaw or Hermes runs the agent and its tools The language model handles reasoning and responses The provider route delivers that model Changing the model does not automatically add skills grant permissions or change your hosting plan When creating an agent describe its job and review the generated preview Find AI route then Hosted model Choose a provider family pick a model and review its context and cost information before hatching The suggested setup is a starting point make sure it fits the work you actually want done For an existing agent open its workspace go to Config and select Model and Provider Keep Managed by Hatcher to use hosted models and AI Credits Use the search provider filters capability tags and cost range to narrow the list Check the model description context window and current pricing Selecting a model changes the draft It does not save the agent yet Compare Saved model with Draft selection then click Save AI Models Wait for the confirmation and any runtime restart Check the active model in chat before your next task Saved presets help you reuse choices but still require saving the configuration The current Hatcher catalog marks DeepSeek V4 Flash as the default hosted model It is positioned as a fast economical starting point Your agent may already have a different saved or suggested model so check it Default does not mean the strongest model and it does not mean automatic selection of the best model for every task For summaries extraction simple replies and repeated routine work start by evaluating economical options such as DeepSeek V4 Flash or Gemini Flash Lite GPT mini models are candidates for broader everyday work DeepSeek Pro offers another step toward more demanding analysis These are starting points for testing not a universal quality ranking For difficult coding planning or long instructions compare stronger options such as Claude Sonnet Claude Opus GPT's higher capability variants or Gemini Pro Codex and Qwen coding models are also worth evaluating for repository work A more capable model may follow constraints better but can cost more and take longer Always review the result The catalog also includes families such as MiMo MiniMax Kimi GLM Grok Mistral and SERV Compare their individual descriptions and prices A provider name alone does not tell you quality or cost A large context window means more room for information not guaranteed accuracy Partner routes can have different pricing and data policies AI Credits meter hosted usage Language models process tokens chunks of text Input includes your prompt instructions history and tool results sent to the model Output includes generated tokens and where billed reasoning tokens Longer conversations larger files and repeated tool loops can increase usage One user message can trigger several model calls Hatcher's billing reference is one thousand AI Credits per dollar of provider usage Credit pack checkout prices are separate Read both input and output rates usually shown per million tokens Actual charges use reported provider cost when available otherwise an estimate applies Positive model charges round up to whole credits with a one credit minimum Here is an illustrative single call using ten thousand input tokens and two thousand output tokens Using the catalog snapshot DeepSeek V4 Flash works out to about one credit after rounding GPT five point four mini is about seventeen Claude Sonnet four point six is about sixty These are list based estimates not recorded charges or a quality benchmark A cheaper call can become expensive if the agent repeats itself or needs several corrections A stronger model can sometimes finish in fewer steps but savings are not guaranteed Compare the same task instructions tools and output requirements Check correctness completion time and total credits including retries Keep the option that meets your quality target Hosted search research embeddings and media tools can add their own charges With Use your own provider supported model calls are billed to your provider account instead of Hatcher AI Credits Hosted tools may still use credits Review your remaining balance and usage history in Billing an estimate or temporary reservation is not necessarily the final charge Start with an economical model for a small representative task If it misses requirements improve the instructions and test a stronger model Use focused context and a clear output length Save your preferred configuration then monitor usage as work grows You are choosing the right model for a job not buying a guarantee of better answers Open your agent choose Model and Provider review the current prices and save your selection Start with a sensible default Spend more when the task earns it Build and manage your agents at Hatcher dot host