diff --git a/tests/llm/test_chat_api.py b/tests/llm/test_chat_api.py
index 5c81a68d9..25eb43a78 100644
--- a/tests/llm/test_chat_api.py
+++ b/tests/llm/test_chat_api.py
@@ -1,11 +1,19 @@
 import os
+from types import SimpleNamespace
 
+import openai
 import pytest
 
+import agentlab.llm.chat_api as chat_api
+import agentlab.llm.tracking as tracking
 from agentlab.llm.chat_api import (
     AnthropicModelArgs,
     AzureModelArgs,
+    ChatModel,
     OpenAIModelArgs,
+    RetryError,
+    _extract_wait_time,
+    handle_error,
     make_system_message,
     make_user_message,
 )
@@ -89,5 +97,145 @@ def test_api_model_args_anthropic():
     assert "5" in answer.get("content")
 
 
+class _FakeCompletions:
+    def __init__(self, responses):
+        self.responses = list(responses)
+        self.calls = []
+
+    def create(self, **kwargs):
+        self.calls.append(kwargs)
+        response = self.responses.pop(0)
+        if isinstance(response, Exception):
+            raise response
+        return response
+
+
+class _FakeClient:
+    def __init__(self, api_key=None, responses=(), **kwargs):
+        self.api_key = api_key
+        self.chat = SimpleNamespace(completions=_FakeCompletions(responses))
+
+
+def _make_completion(content="hello", prompt_tokens=10, completion_tokens=5):
+    return SimpleNamespace(
+        choices=[SimpleNamespace(message=SimpleNamespace(content=content))],
+        usage=SimpleNamespace(prompt_tokens=prompt_tokens, completion_tokens=completion_tokens),
+    )
+
+
+def _make_chat_model(responses, monkeypatch, input_cost=0.1, output_cost=0.2, max_retry=4):
+    """ChatModel wired to a fake client, with sleeps recorded instead of performed."""
+    sleeps = []
+    monkeypatch.setattr(chat_api.time, "sleep", sleeps.append)
+    model = ChatModel(
+        model_name="fake-model",
+        api_key="fake-key",
+        max_retry=max_retry,
+        min_retry_wait_time=1,
+        client_class=_FakeClient,
+        client_args={"responses": responses},
+        pricing_func=lambda: {
+            "fake-model": {"prompt": str(input_cost), "completion": str(output_cost)}
+        },
+    )
+    return model, sleeps
+
+
+def test_extract_wait_time_uses_parsed_value():
+    message = "Rate limit reached. Please try again in 90.5s. Contact us."
+    assert _extract_wait_time(message, min_retry_wait_time=60) == 90.5
+
+
+def test_extract_wait_time_clamps_to_min():
+    message = "Rate limit reached. Please try again in 2s."
+    assert _extract_wait_time(message, min_retry_wait_time=60) == 60
+
+
+def test_extract_wait_time_without_match():
+    assert _extract_wait_time("Some other error", min_retry_wait_time=42) == 42
+
+
+def test_handle_error_reraises_non_openai_error():
+    error = ValueError("not an openai error")
+    with pytest.raises(ValueError):
+        handle_error(error, itr=0, min_retry_wait_time=1, max_retry=4)
+
+
+def test_handle_error_sleeps_and_returns_error_type(monkeypatch):
+    sleeps = []
+    monkeypatch.setattr(chat_api.time, "sleep", sleeps.append)
+    error = openai.OpenAIError("Rate limit reached. Please try again in 90s.")
+
+    error_type = handle_error(error, itr=0, min_retry_wait_time=1, max_retry=4)
+
+    assert error_type == "Rate limit reached. Please try again in 90s."
+    assert sleeps == [90.0]
+
+
+def test_chat_model_retries_then_succeeds(monkeypatch):
+    responses = [
+        openai.OpenAIError("Rate limit reached. Please try again in 3s."),
+        openai.OpenAIError("Rate limit reached. Please try again in 3s."),
+        _make_completion(content="the answer"),
+    ]
+    model, sleeps = _make_chat_model(responses, monkeypatch)
+
+    answer = model([make_user_message("hi")])
+
+    assert answer.get("content") == "the answer"
+    assert model.success
+    assert model.retries == 3
+    assert len(model.error_types) == 2
+    assert sleeps == [3.0, 3.0]
+
+
+def test_chat_model_raises_retry_error_when_retries_exhausted(monkeypatch):
+    responses = [openai.OpenAIError("boom") for _ in range(3)]
+    model, sleeps = _make_chat_model(responses, monkeypatch, max_retry=3)
+
+    with pytest.raises(RetryError, match="after 3 retries"):
+        model([make_user_message("hi")])
+
+    assert not model.success
+    assert model.retries == 3
+    assert sleeps == [1, 1, 1]
+
+
+def test_chat_model_retries_when_usage_is_missing(monkeypatch):
+    no_usage = _make_completion()
+    no_usage.usage = None
+    model, _ = _make_chat_model([no_usage, _make_completion()], monkeypatch)
+
+    model([make_user_message("hi")])
+
+    assert model.retries == 2
+    assert isinstance(model.error_types[0], str)
+
+
+def test_chat_model_cost_is_tracked(monkeypatch):
+    completion = _make_completion(prompt_tokens=10, completion_tokens=5)
+    model, _ = _make_chat_model([completion], monkeypatch, input_cost=0.1, output_cost=0.2)
+
+    with tracking.set_tracker() as tracker:
+        model([make_user_message("hi")])
+
+    assert tracker.input_tokens == 10
+    assert tracker.output_tokens == 5
+    assert tracker.cost == pytest.approx(10 * 0.1 + 5 * 0.2)
+
+
+def test_chat_model_returns_all_samples(monkeypatch):
+    completion = _make_completion()
+    completion.choices = [
+        SimpleNamespace(message=SimpleNamespace(content="a")),
+        SimpleNamespace(message=SimpleNamespace(content="b")),
+    ]
+    model, _ = _make_chat_model([completion], monkeypatch)
+
+    answers = model([make_user_message("hi")], n_samples=2)
+
+    assert [a.get("content") for a in answers] == ["a", "b"]
+
+
 if __name__ == "__main__":
     test_api_model_args_anthropic()