Find the exact Groq model identifier and check its supported task before changing application code. The catalog combines live specifications with guidance on model families, migration and lifecycle risk.
Model families at Groq
Groq’s catalog separates ordinary hosted models from Compound systems and includes text, speech and safety-oriented offerings. An upstream family name identifies the model lineage, while the Groq endpoint identifies the service running your request. Official documentation.
Choose a candidate by the input and output the task actually requires. A speech transcription workflow, a text-only classifier and a tool-using assistant do not become interchangeable because they appear in one catalog. Check the endpoint and response shape before comparing rates.
Compound can use hosted tools, while ordinary model integrations can define local functions for an application to execute. Decide who should own the tool execution and inspect the resulting cost categories. Official documentation.
For a migration, save a set of difficult examples from the existing application. Check instruction following, extracted field accuracy and tool-argument validity rather than relying on a family reputation. Treat benchmark descriptions as evidence with a scope, not a universal quality label.
All models
| Model | Official identifier | Modalities | Context tokens | Input USD / 1M | Output USD / 1M | Status |
|---|---|---|---|---|---|---|
| GPT OSS 120B | openai/gpt-oss-120b | text | 131,072 | $0.15 | $0.6 | Active |
| GPT OSS 20B | openai/gpt-oss-20b | text | 131,072 | $0.075 | $0.3 | Active |
| Qwen/Qwen3.6-27B | qwen/qwen3.6-27b | text | 131,072 | $0.6 | $3 | Active |
| Qwen/Qwen3.8-27B | qwen/qwen3.8-27b | text | 131,042 | $0.8 | $4 | Active |
| Safety GPT OSS 20B | openai/gpt-oss-safeguard-20b | text | 131,072 | $0.075 | $0.3 | Active |
Last verified · Source ↗
The public catalog distinguishes production and preview availability. Follow the lifecycle notes for a selected identifier before using it in a dependency that is difficult to change. Official documentation.
Open a model’s source when a field is missing. An absent context or output value should trigger verification, not an invented default. Keep the identifier in configuration so a deliberate migration does not require searching through several application files.
Cheapest / strongest / longest-context
Only active candidates with documented values for this ranking are included. Prices retain the tier and deployment condition shown below.
| Model | Provider | Input USD / 1M | Output USD / 1M | Context tokens | Price condition | Cost |
|---|---|---|---|---|---|---|
| GPT OSS 20B | Groq | $0.075 | $0.3 | 131,072 | Standard | Estimate cost |
| Safety GPT OSS 20B | Groq | $0.075 | $0.3 | 131,072 | Standard | Estimate cost |
| GPT OSS 120B | Groq | $0.15 | $0.6 | 131,072 | Standard | Estimate cost |
| Qwen/Qwen3.6-27B | Groq | $0.6 | $3 | 131,072 | Standard | Estimate cost |
| Qwen/Qwen3.8-27B | Groq | $0.8 | $4 | 131,042 | Standard | Estimate cost |
Last verified · Source ↗
Only active candidates with documented values for this ranking are included. Prices retain the tier and deployment condition shown below.
| Model | Provider | Input USD / 1M | Output USD / 1M | Cached input USD / 1M | Context tokens | Price condition | Cost |
|---|---|---|---|---|---|---|---|
| GPT OSS 120B | Groq | $0.15 | $0.6 | Not documented | 131,072 | Standard | Estimate cost |
| GPT OSS 20B | Groq | $0.075 | $0.3 | Not documented | 131,072 | Standard | Estimate cost |
| Qwen/Qwen3.6-27B | Groq | $0.6 | $3 | Not documented | 131,072 | Standard | Estimate cost |
| Safety GPT OSS 20B | Groq | $0.075 | $0.3 | Not documented | 131,072 | Standard | Estimate cost |
| Qwen/Qwen3.8-27B | Groq | $0.8 | $4 | Not documented | 131,042 | Standard | Estimate cost |
Last verified · Source ↗
The numeric ranking describes documented price or context capacity. It does not establish the strongest model for your application. Use a saved evaluation set, score the properties the task requires and compare the total effort needed to reach a correct result.
A longer context window is useful only if relevant evidence survives retrieval and prompt assembly. Sending everything can obscure the decisive passage. Compare a focused prompt with a larger one before increasing the default input size.
Deprecations and replacements
Groq maintains lifecycle information and a migration guide. A replacement should be checked as a new integration candidate: request fields, output behavior and operating limits can change even when the application’s visible task stays the same. Official documentation.
Maintain a rollback configuration and preserve examples that passed with the previous model. Deploy the candidate to a controlled subset of work, compare failures and keep a record of the decision. If the old endpoint is retiring, make the migration date visible to the team rather than leaving it inside a release note.
Naming and versions
Use identifiers exactly as listed, including namespaces, slashes and version components. Do not convert a display name into a guessed API identifier. Model permissions can also make a correct identifier inaccessible to a particular project. Official documentation.
A named family is not a promise that every hosted variant accepts the same parameters. Keep capability settings with the selected identifier. When revisiting an old snippet, verify that its example model still appears in the current catalog before copying it.
Use the Groq price list for billing conditions and Groq model-error diagnosis for rejected identifiers. The provider hub links the complete operational reference.
Use the AI API cost calculator to turn the model and workload you are considering into an estimate.
Last verified · Source ↗
Frequently asked questions
Why does my request reject a model shown in the public catalog?
Can I replace a display name with a similar identifier?
Does longest context mean best quality?
What is different about Compound?
What should I verify before a migration?
Sources
Last verified · Source ↗