Skip to content
Python

Groq: current production rows with windows, key link, brand mark

Rohit C Prasad4f6665817acf
diff --git a/coworker/providers/matrix.py b/coworker/providers/matrix.py
index 9fdedbfa..8cb9ff10 100644
--- a/coworker/providers/matrix.py
+++ b/coworker/providers/matrix.py
@@ -16,9 +16,8 @@ where the vendor spec wasn't re-checked stay ``None`` — the meter simply hides
showing a made-up denominator. Values entered 2026-07-28 from vendor docs; verify alongside
the id refresh.
-Resellers: Together + Fireworks + OpenRouter + Nebius + OrcaRouter + NVIDIA NIM. TODO: add Groq entries here AND its
-descriptor in ``registry.py`` once the current provider surface is tested — deliberately
-deferred to bound how much needs verifying at once.
+Resellers: Together + Fireworks + OpenRouter + Nebius + OrcaRouter + NVIDIA NIM
++ Groq (landed 2026-10-25, after the deferral of 2026-07-04).
"""
from __future__ import annotations
@@ -443,13 +442,14 @@ MATRIX: dict[str, ModelEntry] = {
"DeepSeek V4 Pro · via Nebius", _AGENTIC, 1_048_576
),
# -- Groq (LPU inference) ---------------------------------------------------
- "groq:llama-4-maverick-17b-128e-instruct": ModelEntry(
- "Llama 4 Maverick · via Groq"
- ),
- "groq:deepseek-r1-distill-llama-70b": ModelEntry(
- "DeepSeek R1 Distill 70B · via Groq"
- ),
- "groq:qwen-qwq-32b": ModelEntry("Qwen QwQ 32B · via Groq"),
+ # Groq keeps its own short ids. Rows and windows from console.groq.com/docs/models
+ # (production table, read 2026-10-25: 131,072 context each; the earlier Maverick,
+ # R1-distill and QwQ ids are no longer listed). Tool use per console.groq.com/docs/tool-use.
+ "groq:llama-3.3-70b-versatile": ModelEntry(
+ "Llama 3.3 70B Versatile · via Groq", _AGENTIC, 131_072
+ ),
+ "groq:openai/gpt-oss-120b": ModelEntry("GPT-OSS 120B · via Groq", _AGENTIC, 131_072),
+ "groq:openai/gpt-oss-20b": ModelEntry("GPT-OSS 20B · via Groq", _AGENTIC, 131_072),
}
diff --git a/coworker/providers/registry.py b/coworker/providers/registry.py
index aa626eff..c5ad26ad 100644
--- a/coworker/providers/registry.py
+++ b/coworker/providers/registry.py
@@ -718,9 +718,8 @@ DESCRIPTORS: list[ProviderDescriptor] = [
endpoint_help="Prefilled with Novita's OpenAI-compatible endpoint; keep it unless you use a regional or proxy variant.",
),
# Resellers: many labs' models behind one key, using THEIR model namespaces (the curated
- # ids + display labels live in providers/matrix.py). TODO: add Groq here (+ its matrix
- # rows) once the current provider surface is tested — deliberately deferred to bound
- # how much needs verifying at once (owner call, 2026-07-04).
+ # ids + display labels live in providers/matrix.py). Groq was deferred on 2026-07-04 to
+ # bound how much needed verifying at once; it landed 2026-10-25 with verified rows.
_compat(
"together",
"Together AI",
@@ -784,7 +783,7 @@ DESCRIPTORS: list[ProviderDescriptor] = [
"groq",
"Groq",
base_url="https://api.groq.com/openai/v1",
- recommended_model="llama-4-maverick-17b-128e-instruct",
+ recommended_model="llama-3.3-70b-versatile",
env_key="GROQ_API_KEY",
),
ProviderDescriptor(
diff --git a/surfaces/gui/src/providers/ProviderSetup.tsx b/surfaces/gui/src/providers/ProviderSetup.tsx
index 1252b9da..01933ecd 100644
--- a/surfaces/gui/src/providers/ProviderSetup.tsx
+++ b/surfaces/gui/src/providers/ProviderSetup.tsx
@@ -53,6 +53,7 @@ export const KEY_HELP: Record<string, { url: string; label: string; labelKey?: s
xai: { url: "https://console.x.ai", label: "console.x.ai" },
sambanova: { url: "https://cloud.sambanova.ai/apis", label: "cloud.sambanova.ai" },
atlascloud: { url: "https://www.atlascloud.ai/console/api-keys", label: "atlascloud.ai" },
+ groq: { url: "https://console.groq.com/keys", label: "console.groq.com" },
};
export type Verify = { state: "idle" | "testing" | "ok" | "error"; msg?: string };
diff --git a/surfaces/gui/src/providers/logos.ts b/surfaces/gui/src/providers/logos.ts
index db653cd3..3fd1d653 100644
--- a/surfaces/gui/src/providers/logos.ts
+++ b/surfaces/gui/src/providers/logos.ts
@@ -24,6 +24,7 @@ import nvidia from "./logos/nvidia.svg";
import sambanova from "./logos/sambanova.svg";
import atlascloud from "./logos/atlascloud.svg";
import novita from "./logos/novita.svg";
+import groq from "./logos/groq.svg";
import zai from "./logos/zai.svg";
import kimi from "./logos/kimi.svg";
import deepseek from "./logos/deepseek.svg";
@@ -60,6 +61,7 @@ export const PROVIDER_LOGOS: Record<string, string> = {
sambanova,
atlascloud,
novita,
+ groq,
zai,
kimi,
deepseek,
@@ -98,6 +100,7 @@ export const PROVIDER_ORDER = [
"sambanova",
"atlascloud",
"novita",
+ "groq",
];
export function providerRank(name: string): number {
diff --git a/surfaces/gui/src/providers/logos/groq.svg b/surfaces/gui/src/providers/logos/groq.svg
new file mode 100644
index 00000000..7294646b
--- /dev/null
+++ b/surfaces/gui/src/providers/logos/groq.svg
@@ -0,0 +1 @@
+<svg fill="currentColor" fill-rule="evenodd" height="1em" style="flex:none;line-height:1" viewBox="0 0 24 24" width="1em" xmlns="http://www.w3.org/2000/svg"><title>Groq</title><path d="M12.036 2c-3.853-.035-7 3-7.036 6.781-.035 3.782 3.055 6.872 6.908 6.907h2.42v-2.566h-2.292c-2.407.028-4.38-1.866-4.408-4.23-.029-2.362 1.901-4.298 4.308-4.326h.1c2.407 0 4.358 1.915 4.365 4.278v6.305c0 2.342-1.944 4.25-4.323 4.279a4.375 4.375 0 01-3.033-1.252l-1.851 1.818A7 7 0 0012.029 22h.092c3.803-.056 6.858-3.083 6.879-6.816v-6.5C18.907 4.963 15.817 2 12.036 2z"></path></svg>
\ No newline at end of file
diff --git a/tests/test_providers.py b/tests/test_providers.py
index e887b1c6..b82e3b33 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -508,7 +508,7 @@ def test_matrix_answers_capabilities_for_reseller_ids():
# NIM ids are `vendor/model`, so NVIDIA's own models repeat the word in the
# routed id. Deliberate: the `z-ai/` and `deepseek-ai/` rows in matrix.py don't.
"nvidia:nvidia/nemotron-3-super-120b-a12b",
- "groq:llama-4-maverick-17b-128e-instruct",
+ "groq:llama-3.3-70b-versatile",
):
caps = capabilities_for(mid)
assert caps.tools and caps.parallel_tool_calls and caps.streaming