Back to models
Enterprise accessChat Models

Groq

Qwen3.6 27B

Groq-hosted text and vision model with thinking modes and 131K context. Self-service access ended; an enterprise exception remains.

MultimodalThinking modes
Open AppBrowse All Models

At a glance

Know the model before you prompt.

Groq API specifications
Context window
131K tokens (Groq)
Maximum output
Not verified
Inputs → output
Text + Images → Text
Knowledge cutoff
Not verified

API model ID: qwen/qwen3.6-27b

Capabilities & boundaries

What it supports. Where the limits are.

Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.

Supported API features and tools

  • A 27B dense model with a hybrid Gated DeltaNet and Gated Attention architecture described by Groq
  • Text and image understanding for screenshot, document and visual-question tasks on the documented Groq path
  • Thinking and non-thinking modes for balancing deliberation with straightforward dialogue
  • Tool-oriented coding workflows and a published 131K context on Groq, subject to current account access

Before you choose

  • Groq self-service access ended; continued access depends on the stated committed-spend enterprise exception and your actual account.
  • The 262,144-token context on OpenRouter is not the Groq 131K context. OpenRouter's output ceiling and pricing are not verified Groq limits or rates.
  • No exact Groq maximum output, training cutoff or current public token tariff was established in the reviewed source. These remain unverified rather than copied from another host.
  • Do not transfer Qwen3.8's newer controls or structured-output support to Qwen3.6. A successor's documentation is not proof of this endpoint's behavior.

Choose the reasoning effort

nonedefault

Groq's model-specific Qwen3.6 page describes reasoning_effort=default for thinking and none for non-thinking. Here default is a parameter value, not a claim about the default of an omitted request field. It also documents hidden or parsed reasoning_format. Do not import Qwen3.8's additional effort levels.

  • The catalog route stays under Groq. The supplied OpenRouter page is included as a separately hosted reference, not used to fill missing Groq specifications.
  • Groq's Qwen3.6 guidance says to retain final outputs rather than thinking content in multi-turn history. Follow the target endpoint's documented message format.
  • For a migration, consult Groq's current Qwen3.8 documentation and re-test modes, tools and parsing. This guide does not add or promise that successor in EZ Ai Assist.

Put it to work

Start with a more useful prompt.

Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.

Workflow 01

Turn a screenshot into a repair checklist

Combine visual observations with explicit acceptance criteria.

Inspect the attached interface screenshot and the intended user task. List visible obstacles, label each with the visual evidence that supports it, and propose a minimal repair. Separate observations from assumptions about interactions. End with keyboard, mobile and validation checks that a developer should run; do not claim those checks have already passed.

Workflow 02

Read a chart with uncertainty

Avoid false precision when extracting information from an image.

Read this chart and summarize the trend, axes, units and clearly visible values. Mark any unreadable labels or approximate estimates explicitly. Distinguish correlation from explanations that would need outside evidence. Suggest one question the chart cannot answer and what additional data would be needed to resolve it.

Workflow 03

Compare thinking and dialogue modes

Evaluate two modes against the same acceptance criteria.

Using the supplied task examples and scoring rules, prepare a comparison of a thinking mode and a non-thinking mode. Define when a short direct answer is sufficient, when a verification pass is needed and how to measure errors, latency and cost. Provide a blank results table and escalation rules, without inventing outputs or measurements.

Developer reference

Groq API pricing

These are Groq API reference prices, not EZ Ai Assist subscription prices.

View EZ Ai Assist plans

A current public Groq token tariff was not verified for this enterprise-restricted endpoint. Confirm eligible contract pricing; the OpenRouter prices belong to other serving routes and are not substituted here.

  • OpenRouter's supplied Qwen3.6 page is useful for comparing other hosts, but provider rates, limits and access must be evaluated together.
  • Do not carry forward a Preview badge, token-speed estimate or historical rate as evidence of current self-service availability on Groq.

Common questions

A few things worth knowing.

Is Qwen3.6 still available through Groq?

Free and developer-tier access ended September 14, 2026. The deprecation notice exempts enterprise customers with a committed-spend contract. Confirm account eligibility and current app access before use.

Why does OpenRouter show a different context size?

It describes different hosting routes. Groq's model-specific page publishes 131K context; OpenRouter lists 262,144. This guide keeps the hosting boundaries explicit rather than presenting the larger figure as a Groq guarantee.

Can it understand images?

Groq's model page documents text and image understanding, including chart and document tasks. This does not imply image generation or establish that every app workspace exposes image input for this model.

How are thinking and non-thinking selected?

The Qwen3.6 Groq page describes reasoning_effort values default and none respectively. The word default is a literal parameter value here; this guide does not infer the behavior of an omitted field or borrow settings from Qwen3.8.

What replaced it for Groq self-service users?

Groq's notice names Qwen3.8 27B. Review the current successor documentation and test your integration; its recommendation does not mean a new Qwen3.8 page or app integration has been added to this website.

Why are output limits and prices not filled from OpenRouter?

That would mix serving providers. The reviewed Groq source did not verify those current values, so they remain explicitly unverified or contract-dependent. Check the endpoint you actually intend to use.

What should I verify before migrating?

Test both modes, image handling, tool arguments, response parsing, errors and latency on representative tasks. Confirm the new endpoint's limits and rates, and preserve a human review path for uncertain visual interpretations.

Check the source

Official documentation

Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.

Same provider

More from Groq

Groq

GPT-OSS 120B

OpenAI's larger open-weight reasoning model on Groq, with tool use, structured outputs, and a 131K context window.

Groq

GPT-OSS 20B

Compact open-weight reasoning on Groq for focused text tasks, with structured outputs, tool use, and automatic caching.

Groq

Compound Mini

Historical single-tool Groq system retired September 21, 2026. Explore its former workflow and migration considerations.

Groq

Compound

Historical multi-tool Groq system retired September 21, 2026, with no direct replacement listed in its shutdown notice.