pydantic-ai 0.0.20__tar.gz → 0.0.21__tar.gz

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pydantic-ai
-Version: 0.0.20
+Version: 0.0.21
 Summary: Agent Framework / shim to use Pydantic with LLMs
 Project-URL: Homepage, https://ai.pydantic.dev
 Project-URL: Source, https://github.com/pydantic/pydantic-ai
@@ -32,9 +32,9 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: Topic :: Internet
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.9
-Requires-Dist: pydantic-ai-slim[anthropic,cohere,graph,groq,mistral,openai,vertexai]==0.0.20
+Requires-Dist: pydantic-ai-slim[anthropic,cohere,graph,groq,mistral,openai,vertexai]==0.0.21
 Provides-Extra: examples
-Requires-Dist: pydantic-ai-examples==0.0.20; extra == 'examples'
+Requires-Dist: pydantic-ai-examples==0.0.21; extra == 'examples'
 Provides-Extra: logfire
 Requires-Dist: logfire>=2.3; extra == 'logfire'
 Description-Content-Type: text/markdown
@@ -78,7 +78,7 @@ We built PydanticAI with one simple aim: to bring that FastAPI feeling to GenAI
 Built by the team behind [Pydantic](https://docs.pydantic.dev/latest/) (the validation layer of the OpenAI SDK, the Anthropic SDK, LangChain, LlamaIndex, AutoGPT, Transformers, CrewAI, Instructor and many more).
 
 * __Model-agnostic__
-Supports OpenAI, Anthropic, Gemini, Ollama, Groq, and Mistral, and there is a simple interface to implement support for [other models](https://ai.pydantic.dev/models/).
+Supports OpenAI, Anthropic, Gemini, Deepseek, Ollama, Groq, Cohere, and Mistral, and there is a simple interface to implement support for [other models](https://ai.pydantic.dev/models/).
 
 * __Pydantic Logfire Integration__
 Seamlessly [integrates](https://ai.pydantic.dev/logfire/) with [Pydantic Logfire](https://pydantic.dev/logfire) for real-time debugging, performance monitoring, and behavior tracking of your LLM-powered applications.
@@ -116,7 +116,7 @@ from pydantic_ai import Agent
 
 # Define a very simple agent including the model to use, you can also set the model when running the agent.
 agent = Agent(
-    'gemini-1.5-flash',
+    'google-gla:gemini-1.5-flash',
     # Register a static system prompt using a keyword argument to the agent.
     # For more complex dynamically-generated system prompts, see the example below.
     system_prompt='Be concise, reply with one sentence.',

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/README.md

@@ -37,7 +37,7 @@ We built PydanticAI with one simple aim: to bring that FastAPI feeling to GenAI
 Built by the team behind [Pydantic](https://docs.pydantic.dev/latest/) (the validation layer of the OpenAI SDK, the Anthropic SDK, LangChain, LlamaIndex, AutoGPT, Transformers, CrewAI, Instructor and many more).
 
 * __Model-agnostic__
-Supports OpenAI, Anthropic, Gemini, Ollama, Groq, and Mistral, and there is a simple interface to implement support for [other models](https://ai.pydantic.dev/models/).
+Supports OpenAI, Anthropic, Gemini, Deepseek, Ollama, Groq, Cohere, and Mistral, and there is a simple interface to implement support for [other models](https://ai.pydantic.dev/models/).
 
 * __Pydantic Logfire Integration__
 Seamlessly [integrates](https://ai.pydantic.dev/logfire/) with [Pydantic Logfire](https://pydantic.dev/logfire) for real-time debugging, performance monitoring, and behavior tracking of your LLM-powered applications.
@@ -75,7 +75,7 @@ from pydantic_ai import Agent
 
 # Define a very simple agent including the model to use, you can also set the model when running the agent.
 agent = Agent(
-    'gemini-1.5-flash',
+    'google-gla:gemini-1.5-flash',
     # Register a static system prompt using a keyword argument to the agent.
     # For more complex dynamically-generated system prompts, see the example below.
     system_prompt='Be concise, reply with one sentence.',

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "pydantic-ai"
-version = "0.0.20"
+version = "0.0.21"
 description = "Agent Framework / shim to use Pydantic with LLMs"
 authors = [
     { name = "Samuel Colvin", email = "samuel@pydantic.dev" },
@@ -37,7 +37,7 @@ classifiers = [
 ]
 requires-python = ">=3.9"
 
-dependencies = ["pydantic-ai-slim[graph,openai,vertexai,groq,anthropic,mistral,cohere]==0.0.20"]
+dependencies = ["pydantic-ai-slim[graph,openai,vertexai,groq,anthropic,mistral,cohere]==0.0.21"]
 
 [project.urls]
 Homepage = "https://ai.pydantic.dev"
@@ -46,7 +46,7 @@ Documentation = "https://ai.pydantic.dev"
 Changelog = "https://github.com/pydantic/pydantic-ai/releases"
 
 [project.optional-dependencies]
-examples = ["pydantic-ai-examples==0.0.20"]
+examples = ["pydantic-ai-examples==0.0.21"]
 logfire = ["logfire>=2.3"]
 
 [tool.uv.sources]
@@ -183,3 +183,11 @@ ignore_no_config = true
 [tool.inline-snapshot.shortcuts]
 snap-fix=["create", "fix"]
 snap=["create"]
+
+[tool.codespell]
+# Ref: https://github.com/codespell-project/codespell#using-a-config-file
+skip = '.git*,*.svg,*.lock,*.css'
+check-hidden = true
+# Ignore "formatting" like **L**anguage
+ignore-regex = '\*\*[A-Z]\*\*[a-z]+\b'
+# ignore-words-list = ''

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/tests/models/test_anthropic.py

@@ -11,7 +11,6 @@ from inline_snapshot import snapshot
 
 from pydantic_ai import Agent, ModelRetry
 from pydantic_ai.messages import (
-    ArgsDict,
     ModelRequest,
     ModelResponse,
     RetryPromptPart,
@@ -47,7 +46,7 @@ with try_import() as imports_successful:
     )
     from anthropic.types.raw_message_delta_event import Delta
 
-    from pydantic_ai.models.anthropic import AnthropicModel
+    from pydantic_ai.models.anthropic import AnthropicModel, AnthropicModelSettings
 
 pytestmark = [
     pytest.mark.skipif(not imports_successful(), reason='anthropic not installed'),
@@ -186,7 +185,7 @@ async def test_request_structured_response(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='final_result',
-                        args=ArgsDict(args_dict={'response': [1, 2, 3]}),
+                        args={'response': [1, 2, 3]},
                         tool_call_id='123',
                     )
                 ],
@@ -248,7 +247,7 @@ async def test_request_tool_call(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsDict(args_dict={'loc_name': 'San Francisco'}),
+                        args={'loc_name': 'San Francisco'},
                         tool_call_id='1',
                     )
                 ],
@@ -269,7 +268,7 @@ async def test_request_tool_call(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsDict(args_dict={'loc_name': 'London'}),
+                        args={'loc_name': 'London'},
                         tool_call_id='2',
                     )
                 ],
@@ -332,6 +331,17 @@ async def test_parallel_tool_calls(allow_model_requests: None, parallel_tool_cal
     )
 
 
+async def test_anthropic_specific_metadata(allow_model_requests: None) -> None:
+    c = completion_message([TextBlock(text='world', type='text')], AnthropicUsage(input_tokens=5, output_tokens=10))
+    mock_client = MockAnthropic.create_mock(c)
+    m = AnthropicModel('claude-3-5-haiku-latest', anthropic_client=mock_client)
+    agent = Agent(m)
+
+    result = await agent.run('hello', model_settings=AnthropicModelSettings(anthropic_metadata={'user_id': '123'}))
+    assert result.data == 'world'
+    assert get_mock_chat_completion_kwargs(mock_client)[0]['metadata']['user_id'] == '123'
+
+
 async def test_stream_structured(allow_model_requests: None):
     """Test streaming structured responses with Anthropic's API.
 

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/tests/models/test_cohere.py

@@ -169,7 +169,7 @@ async def test_request_structured_response(allow_model_requests: None):
             ModelRequest(parts=[UserPromptPart(content='Hello', timestamp=IsNow(tz=timezone.utc))]),
             ModelResponse(
                 parts=[
-                    ToolCallPart.from_raw_args(
+                    ToolCallPart(
                         tool_name='final_result',
                         args='{"response": [1, 2, 123]}',
                         tool_call_id='123',
@@ -255,7 +255,7 @@ async def test_request_tool_call(allow_model_requests: None):
             ),
             ModelResponse(
                 parts=[
-                    ToolCallPart.from_raw_args(
+                    ToolCallPart(
                         tool_name='get_location',
                         args='{"loc_name": "San Fransisco"}',
                         tool_call_id='1',
@@ -276,7 +276,7 @@ async def test_request_tool_call(allow_model_requests: None):
             ),
             ModelResponse(
                 parts=[
-                    ToolCallPart.from_raw_args(
+                    ToolCallPart(
                         tool_name='get_location',
                         args='{"loc_name": "London"}',
                         tool_call_id='2',

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/tests/models/test_gemini.py

@@ -15,7 +15,6 @@ from typing_extensions import Literal, TypeAlias
 
 from pydantic_ai import Agent, ModelRetry, UnexpectedModelBehavior, UserError
 from pydantic_ai.messages import (
-    ArgsDict,
     ModelRequest,
     ModelResponse,
     RetryPromptPart,
@@ -474,9 +473,7 @@ async def test_text_success(get_gemini_client: GetGeminiClient):
 
 async def test_request_structured_response(get_gemini_client: GetGeminiClient):
     response = gemini_response(
-        _content_model_response(
-            ModelResponse(parts=[ToolCallPart.from_raw_args('final_result', {'response': [1, 2, 123]})])
-        )
+        _content_model_response(ModelResponse(parts=[ToolCallPart('final_result', {'response': [1, 2, 123]})]))
     )
     gemini_client = get_gemini_client(response)
     m = GeminiModel('gemini-1.5-flash', http_client=gemini_client)
@@ -491,7 +488,7 @@ async def test_request_structured_response(get_gemini_client: GetGeminiClient):
                 parts=[
                     ToolCallPart(
                         tool_name='final_result',
-                        args=ArgsDict(args_dict={'response': [1, 2, 123]}),
+                        args={'response': [1, 2, 123]},
                     )
                 ],
                 model_name='gemini-1.5-flash',
@@ -511,16 +508,14 @@ async def test_request_structured_response(get_gemini_client: GetGeminiClient):
 async def test_request_tool_call(get_gemini_client: GetGeminiClient):
     responses = [
         gemini_response(
-            _content_model_response(
-                ModelResponse(parts=[ToolCallPart.from_raw_args('get_location', {'loc_name': 'San Fransisco'})])
-            )
+            _content_model_response(ModelResponse(parts=[ToolCallPart('get_location', {'loc_name': 'San Fransisco'})]))
         ),
         gemini_response(
             _content_model_response(
                 ModelResponse(
                     parts=[
-                        ToolCallPart.from_raw_args('get_location', {'loc_name': 'London'}),
-                        ToolCallPart.from_raw_args('get_location', {'loc_name': 'New York'}),
+                        ToolCallPart('get_location', {'loc_name': 'London'}),
+                        ToolCallPart('get_location', {'loc_name': 'New York'}),
                     ]
                 )
             )
@@ -554,7 +549,7 @@ async def test_request_tool_call(get_gemini_client: GetGeminiClient):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsDict(args_dict={'loc_name': 'San Fransisco'}),
+                        args={'loc_name': 'San Fransisco'},
                     )
                 ],
                 model_name='gemini-1.5-flash',
@@ -573,11 +568,11 @@ async def test_request_tool_call(get_gemini_client: GetGeminiClient):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsDict(args_dict={'loc_name': 'London'}),
+                        args={'loc_name': 'London'},
                     ),
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsDict(args_dict={'loc_name': 'New York'}),
+                        args={'loc_name': 'New York'},
                     ),
                 ],
                 model_name='gemini-1.5-flash',
@@ -664,9 +659,7 @@ async def test_stream_text_no_data(get_gemini_client: GetGeminiClient):
 async def test_stream_structured(get_gemini_client: GetGeminiClient):
     responses = [
         gemini_response(
-            _content_model_response(
-                ModelResponse(parts=[ToolCallPart.from_raw_args('final_result', {'response': [1, 2]})])
-            ),
+            _content_model_response(ModelResponse(parts=[ToolCallPart('final_result', {'response': [1, 2]})])),
         ),
     ]
     json_data = _gemini_streamed_response_ta.dump_json(responses, by_alias=True)
@@ -684,10 +677,10 @@ async def test_stream_structured(get_gemini_client: GetGeminiClient):
 async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
     first_responses = [
         gemini_response(
-            _content_model_response(ModelResponse(parts=[ToolCallPart.from_raw_args('foo', {'x': 'a'})])),
+            _content_model_response(ModelResponse(parts=[ToolCallPart('foo', {'x': 'a'})])),
         ),
         gemini_response(
-            _content_model_response(ModelResponse(parts=[ToolCallPart.from_raw_args('bar', {'y': 'b'})])),
+            _content_model_response(ModelResponse(parts=[ToolCallPart('bar', {'y': 'b'})])),
         ),
     ]
     d1 = _gemini_streamed_response_ta.dump_json(first_responses, by_alias=True)
@@ -695,9 +688,7 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
 
     second_responses = [
         gemini_response(
-            _content_model_response(
-                ModelResponse(parts=[ToolCallPart.from_raw_args('final_result', {'response': [1, 2]})])
-            ),
+            _content_model_response(ModelResponse(parts=[ToolCallPart('final_result', {'response': [1, 2]})])),
         ),
     ]
     d2 = _gemini_streamed_response_ta.dump_json(second_responses, by_alias=True)
@@ -727,8 +718,8 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
             ModelRequest(parts=[UserPromptPart(content='Hello', timestamp=IsNow(tz=timezone.utc))]),
             ModelResponse(
                 parts=[
-                    ToolCallPart(tool_name='foo', args=ArgsDict(args_dict={'x': 'a'})),
-                    ToolCallPart(tool_name='bar', args=ArgsDict(args_dict={'y': 'b'})),
+                    ToolCallPart(tool_name='foo', args={'x': 'a'}),
+                    ToolCallPart(tool_name='bar', args={'y': 'b'}),
                 ],
                 model_name='gemini-1.5-flash',
                 timestamp=IsNow(tz=timezone.utc),
@@ -743,7 +734,7 @@ async def test_stream_structured_tool_calls(get_gemini_client: GetGeminiClient):
                 parts=[
                     ToolCallPart(
                         tool_name='final_result',
-                        args=ArgsDict(args_dict={'response': [1, 2]}),
+                        args={'response': [1, 2]},
                     )
                 ],
                 model_name='gemini-1.5-flash',
@@ -772,7 +763,7 @@ async def test_stream_text_heterogeneous(get_gemini_client: GetGeminiClient):
                     _function_call_part_from_call(
                         ToolCallPart(
                             tool_name='get_location',
-                            args=ArgsDict(args_dict={'loc_name': 'San Fransisco'}),
+                            args={'loc_name': 'San Fransisco'},
                         )
                     ),
                 ],
@@ -795,7 +786,7 @@ async def test_empty_text_ignored():
     content = _content_model_response(
         ModelResponse(
             parts=[
-                ToolCallPart.from_raw_args('final_result', {'response': [1, 2, 123]}),
+                ToolCallPart('final_result', {'response': [1, 2, 123]}),
                 TextPart(content='xxx'),
             ]
         )
@@ -814,7 +805,7 @@ async def test_empty_text_ignored():
     content = _content_model_response(
         ModelResponse(
             parts=[
-                ToolCallPart.from_raw_args('final_result', {'response': [1, 2, 123]}),
+                ToolCallPart('final_result', {'response': [1, 2, 123]}),
                 TextPart(content=''),
             ]
         )
@@ -826,3 +817,39 @@ async def test_empty_text_ignored():
             'parts': [{'function_call': {'name': 'final_result', 'args': {'response': [1, 2, 123]}}}],
         }
     )
+
+
+async def test_model_settings(client_with_handler: ClientWithHandler, env: TestEnv, allow_model_requests: None) -> None:
+    def handler(request: httpx.Request) -> httpx.Response:
+        generation_config = json.loads(request.content)['generation_config']
+        assert generation_config == {
+            'max_output_tokens': 1,
+            'temperature': 0.1,
+            'top_p': 0.2,
+            'presence_penalty': 0.3,
+            'frequency_penalty': 0.4,
+        }
+        return httpx.Response(
+            200,
+            content=_gemini_response_ta.dump_json(
+                gemini_response(_content_model_response(ModelResponse(parts=[TextPart('world')]))),
+                by_alias=True,
+            ),
+            headers={'Content-Type': 'application/json'},
+        )
+
+    gemini_client = client_with_handler(handler)
+    m = GeminiModel('gemini-1.5-flash', http_client=gemini_client, api_key='mock')
+    agent = Agent(m)
+
+    result = await agent.run(
+        'hello',
+        model_settings={
+            'max_tokens': 1,
+            'temperature': 0.1,
+            'top_p': 0.2,
+            'presence_penalty': 0.3,
+            'frequency_penalty': 0.4,
+        },
+    )
+    assert result.data == 'world'

{pydantic_ai-0.0.20 → pydantic_ai-0.0.21}/tests/models/test_groq.py

@@ -13,7 +13,6 @@ from typing_extensions import TypedDict
 
 from pydantic_ai import Agent, ModelRetry, UnexpectedModelBehavior
 from pydantic_ai.messages import (
-    ArgsJson,
     ModelRequest,
     ModelResponse,
     RetryPromptPart,
@@ -51,9 +50,9 @@ pytestmark = [
 
 
 def test_init():
-    m = GroqModel('llama-3.1-70b-versatile', api_key='foobar')
+    m = GroqModel('llama-3.3-70b-versatile', api_key='foobar')
     assert m.client.api_key == 'foobar'
-    assert m.name() == 'groq:llama-3.1-70b-versatile'
+    assert m.name() == 'groq:llama-3.3-70b-versatile'
 
 
 @dataclass
@@ -103,7 +102,7 @@ def completion_message(message: ChatCompletionMessage, *, usage: CompletionUsage
         id='123',
         choices=[Choice(finish_reason='stop', index=0, message=message)],
         created=1704067200,  # 2024-01-01
-        model='llama-3.1-70b-versatile',
+        model='llama-3.3-70b-versatile',
         object='chat.completion',
         usage=usage,
     )
@@ -112,7 +111,7 @@ def completion_message(message: ChatCompletionMessage, *, usage: CompletionUsage
 async def test_request_simple_success(allow_model_requests: None):
     c = completion_message(ChatCompletionMessage(content='world', role='assistant'))
     mock_client = MockGroq.create_mock(c)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m)
 
     result = await agent.run('hello')
@@ -130,13 +129,13 @@ async def test_request_simple_success(allow_model_requests: None):
             ModelRequest(parts=[UserPromptPart(content='hello', timestamp=IsNow(tz=timezone.utc))]),
             ModelResponse(
                 parts=[TextPart(content='world')],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
             ModelRequest(parts=[UserPromptPart(content='hello', timestamp=IsNow(tz=timezone.utc))]),
             ModelResponse(
                 parts=[TextPart(content='world')],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
         ]
@@ -149,7 +148,7 @@ async def test_request_simple_usage(allow_model_requests: None):
         usage=CompletionUsage(completion_tokens=1, prompt_tokens=2, total_tokens=3),
     )
     mock_client = MockGroq.create_mock(c)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m)
 
     result = await agent.run('Hello')
@@ -171,7 +170,7 @@ async def test_request_structured_response(allow_model_requests: None):
         )
     )
     mock_client = MockGroq.create_mock(c)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m, result_type=list[int])
 
     result = await agent.run('Hello')
@@ -183,11 +182,11 @@ async def test_request_structured_response(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='final_result',
-                        args=ArgsJson(args_json='{"response": [1, 2, 123]}'),
+                        args='{"response": [1, 2, 123]}',
                         tool_call_id='123',
                     )
                 ],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, tzinfo=timezone.utc),
             ),
             ModelRequest(
@@ -245,7 +244,7 @@ async def test_request_tool_call(allow_model_requests: None):
         completion_message(ChatCompletionMessage(content='final response', role='assistant')),
     ]
     mock_client = MockGroq.create_mock(responses)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m, system_prompt='this is the system prompt')
 
     @agent.tool_plain
@@ -269,11 +268,11 @@ async def test_request_tool_call(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsJson(args_json='{"loc_name": "San Fransisco"}'),
+                        args='{"loc_name": "San Fransisco"}',
                         tool_call_id='1',
                     )
                 ],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
             ModelRequest(
@@ -290,11 +289,11 @@ async def test_request_tool_call(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='get_location',
-                        args=ArgsJson(args_json='{"loc_name": "London"}'),
+                        args='{"loc_name": "London"}',
                         tool_call_id='2',
                     )
                 ],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
             ModelRequest(
@@ -309,7 +308,7 @@ async def test_request_tool_call(allow_model_requests: None):
             ),
             ModelResponse(
                 parts=[TextPart(content='final response')],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
         ]
@@ -327,7 +326,7 @@ def chunk(delta: list[ChoiceDelta], finish_reason: FinishReason | None = None) -
         ],
         created=1704067200,  # 2024-01-01
         x_groq=None,
-        model='llama-3.1-70b-versatile',
+        model='llama-3.3-70b-versatile',
         object='chat.completion.chunk',
         usage=CompletionUsage(completion_tokens=1, prompt_tokens=2, total_tokens=3),
     )
@@ -340,7 +339,7 @@ def text_chunk(text: str, finish_reason: FinishReason | None = None) -> chat.Cha
 async def test_stream_text(allow_model_requests: None):
     stream = text_chunk('hello '), text_chunk('world'), chunk([])
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m)
 
     async with agent.run_stream('') as result:
@@ -352,7 +351,7 @@ async def test_stream_text(allow_model_requests: None):
 async def test_stream_text_finish_reason(allow_model_requests: None):
     stream = text_chunk('hello '), text_chunk('world'), text_chunk('.', finish_reason='stop')
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m)
 
     async with agent.run_stream('') as result:
@@ -399,7 +398,7 @@ async def test_stream_structured(allow_model_requests: None):
         chunk([]),
     )
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m, result_type=MyTypedDict)
 
     async with agent.run_stream('') as result:
@@ -422,10 +421,10 @@ async def test_stream_structured(allow_model_requests: None):
                 parts=[
                     ToolCallPart(
                         tool_name='final_result',
-                        args=ArgsJson(args_json='{"first": "One", "second": "Two"}'),
+                        args='{"first": "One", "second": "Two"}',
                     )
                 ],
-                model_name='llama-3.1-70b-versatile',
+                model_name='llama-3.3-70b-versatile',
                 timestamp=datetime(2024, 1, 1, 0, 0, tzinfo=timezone.utc),
             ),
             ModelRequest(
@@ -450,7 +449,7 @@ async def test_stream_structured_finish_reason(allow_model_requests: None):
         struc_chunk(None, None, finish_reason='stop'),
     )
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m, result_type=MyTypedDict)
 
     async with agent.run_stream('') as result:
@@ -470,7 +469,7 @@ async def test_stream_structured_finish_reason(allow_model_requests: None):
 async def test_no_content(allow_model_requests: None):
     stream = chunk([ChoiceDelta()]), chunk([ChoiceDelta()])
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m, result_type=MyTypedDict)
 
     with pytest.raises(UnexpectedModelBehavior, match='Received empty model response'):
@@ -481,7 +480,7 @@ async def test_no_content(allow_model_requests: None):
 async def test_no_delta(allow_model_requests: None):
     stream = chunk([]), text_chunk('hello '), text_chunk('world')
     mock_client = MockGroq.create_mock_stream(stream)
-    m = GroqModel('llama-3.1-70b-versatile', groq_client=mock_client)
+    m = GroqModel('llama-3.3-70b-versatile', groq_client=mock_client)
     agent = Agent(m)
 
     async with agent.run_stream('') as result: