Skip to content
Python

Add GLM 5.3 via OpenRouter to the model matrix, sized for the Z.AI host (OPE-215)

Devika Verma8f5d4045824f
diff --git a/coworker/providers/matrix.py b/coworker/providers/matrix.py
index de2e1086..ec2917ea 100644
--- a/coworker/providers/matrix.py
+++ b/coworker/providers/matrix.py
@@ -278,6 +278,14 @@ MATRIX: dict[str, ModelEntry] = {
"openrouter:nvidia/nemotron-3.5-lightning": ModelEntry(
"Nemotron 3.5 Lightning · via OpenRouter", _AGENTIC, 262_144
),
+ # GLM 5.3 via OpenRouter (openrouter.ai/api/v1/models/z-ai/glm-5.3/endpoints, read
+ # 2026-10-01: 39 endpoints, all tool-capable). Sized for Z.AI, the host a pinned run
+ # uses (1,048,576 ctx, fp8, 131,072 max output). Without this row the model would fall
+ # back to DEFAULT_CONTEXT_WINDOW (128,000) and compact at 102,400, far inside what the
+ # endpoint actually serves (OPE-215).
+ "openrouter:z-ai/glm-5.3": ModelEntry(
+ "GLM 5.3 · via OpenRouter", _AGENTIC, 1_048_576
+ ),
# Stealth/cloaked alpha (catalog-checked 2026-08-24: 1,048,576 ctx, tool calling).
# These are temporary lab previews — expect the slug to vanish when the lab ships
# the real model; keep it until OpenRouter retires it.
diff --git a/tests/test_matrix_current_models.py b/tests/test_matrix_current_models.py
index 30f7c375..ea2516da 100644
--- a/tests/test_matrix_current_models.py
+++ b/tests/test_matrix_current_models.py
@@ -54,6 +54,8 @@ REPO = Path(__file__).resolve().parents[1]
# context_length (202,800), not the model-level 262,144 the catalog shows.
("openrouter:nvidia/nemotron-3-ultra-550b-a55b", 202_800),
("openrouter:nvidia/nemotron-3.5-lightning", 262_144),
+ # openrouter.ai endpoints API, read 2026-10-01 (OPE-215): the Z.AI host's window.
+ ("openrouter:z-ai/glm-5.3", 1_048_576),
],
)
def test_current_models_have_verified_windows(model, window):
diff --git a/tests/test_providers.py b/tests/test_providers.py
index 604a49a4..cd0daad7 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -517,7 +517,10 @@ def test_matrix_labels_and_custom_model_fallback():
# pruning owed above is NOT done here — deciding which entries are retired is an owner
# call, and dropping a row silently downgrades that model to the conservative fallback
# capabilities. Prune before raising this a third time.
- assert len(MATRIX) < 70
+ # 70→75 (2026-10-01, OPE-215): three OpenRouter rows (Nemotron 3 Ultra, Nemotron 3.5
+ # Lightning, GLM 5.3). The pruning owed above is still not done; it remains the owner's
+ # call and should come before this is raised again.
+ assert len(MATRIX) < 75
assert all(e.caps.tools for e in MATRIX.values())
# A custom (unlisted) reseller model falls back to the conservative default — usable,
# but at the user's own risk (no parallel tool calls assumed).