Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)
Python
Switch to GPT-5.2 Codex models and upgrade OpenAI SDK
Abi Rajabb13948264c6
diff --git a/backend/agentic/runner.py b/backend/agentic/runner.pyindex 31eaa9c8..489e30da 100644--- a/backend/agentic/runner.py+++ b/backend/agentic/runner.py@@ -13,7 +13,15 @@ from google.genai import typesfrom codegen.utils import extract_html_contentfrom config import REPLICATE_API_KEYfrom image_generation.core import process_tasks-from llm import Llm, OPENAI_MODELS, ANTHROPIC_MODELS, GEMINI_MODELS+from llm import (+ Llm,+ OPENAI_MODELS,+ ANTHROPIC_MODELS,+ GEMINI_MODELS,+ OPENAI_CODEX_MODELS,+ get_openai_api_name,+ get_openai_reasoning_effort,+)@dataclass@@ -63,6 +71,43 @@ def _parse_json_arguments(raw_args: Any) -> Tuple[Dict[str, Any], Optional[str]]return {}, f"Invalid JSON arguments: {exc}"+def _convert_message_to_responses_input(+ message: ChatCompletionMessageParam,+) -> Dict[str, Any]:+ role = message.get("role", "user")+ content = message.get("content", "")++ if isinstance(content, str):+ return {"role": role, "content": content}++ parts: List[Dict[str, Any]] = []+ if isinstance(content, list):+ for part in content:+ if not isinstance(part, dict):+ continue+ if part.get("type") == "text":+ parts.append({"type": "input_text", "text": part.get("text", "")})+ elif part.get("type") == "image_url":+ image_url = part.get("image_url", {})+ parts.append(+ {+ "type": "input_image",+ "image_url": image_url.get("url", ""),+ "detail": image_url.get("detail", "auto"),+ }+ )++ return {"role": role, "content": parts}+++def _get_event_attr(event: Any, key: str, default: Any = None) -> Any:+ if hasattr(event, key):+ return getattr(event, key)+ if isinstance(event, dict):+ return event.get(key, default)+ return default++def _strip_incomplete_escape(value: str) -> str:if not value:return value@@ -670,6 +715,8 @@ class AgenticRunner:async def _run_openai(self, model: Llm, prompt_messages: List[ChatCompletionMessageParam]) -> str:+ if model in OPENAI_CODEX_MODELS:+ return await self._run_openai_responses(model, prompt_messages)if not self.openai_api_key:raise Exception("OpenAI API key is missing.")@@ -686,17 +733,14 @@ class AgenticRunner:started_tool_ids: set[str] = set()params: Dict[str, Any] = {- "model": model.value,+ "model": get_openai_api_name(model),"messages": messages,"tools": openai_tools,"tool_choice": "auto","temperature": 0,"stream": True,}- if model.value == "gpt-5.2-2025-12-11":- params["max_completion_tokens"] = 30000- else:- params["max_tokens"] = 30000+ params["max_tokens"] = 30000stream = await client.chat.completions.create(**params)@@ -766,9 +810,10 @@ class AgenticRunner:args, error = _parse_json_arguments(entry.get("arguments"))if error:args = {"_raw": entry.get("arguments")}+ tool_call_id = entry.get("id") or f"tool-{uuid.uuid4().hex[:6]}"tool_call_list.append(ToolCall(- id=entry["id"],+ id=tool_call_id,name=entry.get("name") or "unknown_tool",arguments=args,)@@ -835,6 +880,180 @@ class AgenticRunner:finally:await client.close()+ async def _run_openai_responses(+ self, model: Llm, prompt_messages: List[ChatCompletionMessageParam]+ ) -> str:+ if not self.openai_api_key:+ raise Exception("OpenAI API key is missing.")++ client = AsyncOpenAI(api_key=self.openai_api_key, base_url=self.openai_base_url)+ openai_tools, _, _ = self._tool_schemas()+ base_input = [_convert_message_to_responses_input(m) for m in prompt_messages]+ previous_response_id: Optional[str] = None++ max_steps = 8+ try:+ for _ in range(max_steps):+ assistant_event_id = self._next_event_id("assistant")+ assistant_text = ""+ tool_calls: Dict[str, Dict[str, Any]] = {}+ response_id: Optional[str] = None++ params: Dict[str, Any] = {+ "model": get_openai_api_name(model),+ "input": base_input,+ "tools": openai_tools,+ "tool_choice": "auto",+ "stream": True,+ "max_output_tokens": 30000,+ }+ reasoning_effort = get_openai_reasoning_effort(model)+ if reasoning_effort:+ params["reasoning"] = {"effort": reasoning_effort}+ if previous_response_id:+ params["previous_response_id"] = previous_response_id++ responses_client = getattr(client, "responses", None)+ if responses_client is None:+ raise Exception(+ "OpenAI SDK is too old for GPT-5.2 Codex. Please upgrade the 'openai' package to a version that supports the Responses API."+ )+ stream = await responses_client.create(**params) # type: ignore++ async for event in stream: # type: ignore+ event_type = _get_event_attr(event, "type")+ if event_type in ("response.created", "response.completed"):+ response = _get_event_attr(event, "response")+ response_id = (+ _get_event_attr(response, "id")+ or _get_event_attr(event, "response_id")+ or response_id+ )+ continue+ if event_type == "response.output_text.delta":+ delta = _get_event_attr(event, "delta", "")+ if delta:+ assistant_text += delta+ await self._send(+ "assistant", delta, event_id=assistant_event_id+ )+ continue+ if event_type == "response.output_item.added":+ item = _get_event_attr(event, "item")+ if item and _get_event_attr(item, "type") == "function_call":+ call_id = _get_event_attr(item, "call_id") or _get_event_attr(+ item, "id"+ )+ if call_id:+ tool_calls[call_id] = {+ "id": call_id,+ "name": _get_event_attr(item, "name"),+ "arguments": _get_event_attr(item, "arguments", ""),+ }+ continue+ if event_type == "response.function_call_arguments.delta":+ call_id = _get_event_attr(event, "call_id") or _get_event_attr(+ event, "item_id"+ )+ if not call_id:+ continue+ entry = tool_calls.setdefault(+ call_id,+ {+ "id": call_id,+ "name": _get_event_attr(event, "name"),+ "arguments": "",+ },+ )+ entry["arguments"] += _get_event_attr(event, "delta", "") or ""+ continue+ if event_type == "response.function_call_arguments.done":+ call_id = _get_event_attr(event, "call_id") or _get_event_attr(+ event, "item_id"+ )+ if not call_id:+ continue+ entry = tool_calls.setdefault(+ call_id,+ {+ "id": call_id,+ "name": _get_event_attr(event, "name"),+ "arguments": "",+ },+ )+ entry["arguments"] = _get_event_attr(+ event, "arguments", entry["arguments"]+ )+ if _get_event_attr(event, "name"):+ entry["name"] = _get_event_attr(event, "name")++ if not tool_calls:+ return await self._finalize_response(assistant_text)++ tool_call_list: List[ToolCall] = []+ for entry in tool_calls.values():+ args, error = _parse_json_arguments(entry.get("arguments"))+ if error:+ args = {"_raw": entry.get("arguments")}+ tool_call_list.append(+ ToolCall(+ id=entry["id"],+ name=entry.get("name") or "unknown_tool",+ arguments=args,+ )+ )++ tool_output_items: List[Dict[str, Any]] = []+ for tool_call in tool_call_list:+ tool_event_id = tool_call.id or self._next_event_id("tool")+ await self._send(+ "toolStart",+ data={+ "name": tool_call.name,+ "input": self._summarize_tool_input(tool_call),+ },+ event_id=tool_event_id,+ )+ if tool_call.name == "create_file":+ content = _extract_content_from_args(tool_call.arguments)+ if content:+ await self._stream_code_preview(tool_event_id, content)++ tool_result = await self.toolbox.execute(tool_call)+ if tool_result.updated_content:+ await self._send("setCode", tool_result.updated_content)++ await self._send(+ "toolResult",+ data={+ "name": tool_call.name,+ "output": tool_result.summary,+ "ok": tool_result.ok,+ },+ event_id=tool_event_id,+ )++ tool_output_items.append(+ {+ "type": "function_call_output",+ "call_id": tool_call.id,+ "output": json.dumps(tool_result.result),+ }+ )++ if response_id:+ previous_response_id = response_id+ base_input = tool_output_items+ else:+ base_input = [+ _convert_message_to_responses_input(m)+ for m in prompt_messages+ ] + tool_output_items++ raise Exception("Agent exceeded max tool turns")+ finally:+ await client.close()+async def _run_anthropic(self, model: Llm, prompt_messages: List[ChatCompletionMessageParam]) -> str:diff --git a/backend/evals/core.py b/backend/evals/core.pyindex e34b349b..394295c5 100644--- a/backend/evals/core.py+++ b/backend/evals/core.py@@ -51,7 +51,7 @@ async def generate_code_core(api_key=OPENAI_API_KEY,base_url=None,callback=lambda x: process_chunk(x),- model_name=model.value,+ model=model,)return completion["code"]diff --git a/backend/llm.py b/backend/llm.pyindex 5b7742f0..0860b407 100644--- a/backend/llm.py+++ b/backend/llm.py@@ -6,7 +6,8 @@ from typing import TypedDictclass Llm(Enum):# GPTGPT_4_1_2025_04_14 = "gpt-4.1-2025-04-14"- GPT_5_2_2025_12_11 = "gpt-5.2-2025-12-11"+ GPT_5_2_CODEX_MEDIUM = "gpt-5.2-codex (medium thinking)"+ GPT_5_2_CODEX_HIGH = "gpt-5.2-codex (high thinking)"# ClaudeCLAUDE_4_5_SONNET_2025_09_29 = "claude-sonnet-4-5-20250929"CLAUDE_4_5_OPUS_2025_11_01 = "claude-opus-4-5-20251101"@@ -28,7 +29,8 @@ class Completion(TypedDict):MODEL_PROVIDER: dict[Llm, str] = {# OpenAI modelsLlm.GPT_4_1_2025_04_14: "openai",- Llm.GPT_5_2_2025_12_11: "openai",+ Llm.GPT_5_2_CODEX_MEDIUM: "openai",+ Llm.GPT_5_2_CODEX_HIGH: "openai",# Anthropic modelsLlm.CLAUDE_4_5_SONNET_2025_09_29: "anthropic",Llm.CLAUDE_4_5_OPUS_2025_11_01: "anthropic",@@ -43,3 +45,23 @@ MODEL_PROVIDER: dict[Llm, str] = {OPENAI_MODELS = {m for m, p in MODEL_PROVIDER.items() if p == "openai"}ANTHROPIC_MODELS = {m for m, p in MODEL_PROVIDER.items() if p == "anthropic"}GEMINI_MODELS = {m for m, p in MODEL_PROVIDER.items() if p == "gemini"}++OPENAI_MODEL_CONFIG: dict[Llm, dict[str, str]] = {+ Llm.GPT_4_1_2025_04_14: {"api_name": "gpt-4.1-2025-04-14"},+ Llm.GPT_5_2_CODEX_MEDIUM: {"api_name": "gpt-5.2-codex", "reasoning_effort": "medium"},+ Llm.GPT_5_2_CODEX_HIGH: {"api_name": "gpt-5.2-codex", "reasoning_effort": "high"},+}+++def get_openai_api_name(model: Llm) -> str:+ return OPENAI_MODEL_CONFIG[model]["api_name"]+++def get_openai_reasoning_effort(model: Llm) -> str | None:+ return OPENAI_MODEL_CONFIG.get(model, {}).get("reasoning_effort")+++OPENAI_CODEX_MODELS = {+ Llm.GPT_5_2_CODEX_MEDIUM,+ Llm.GPT_5_2_CODEX_HIGH,+}diff --git a/backend/models/openai_client.py b/backend/models/openai_client.pyindex 616b4eb9..f40e51fa 100644--- a/backend/models/openai_client.py+++ b/backend/models/openai_client.py@@ -1,8 +1,37 @@import time-from typing import Awaitable, Callable, List+from typing import Awaitable, Callable, List, Any, Dictfrom openai import AsyncOpenAIfrom openai.types.chat import ChatCompletionMessageParam, ChatCompletionChunk-from llm import Completion+from llm import Completion, Llm, get_openai_api_name, get_openai_reasoning_effort, OPENAI_CODEX_MODELS+++def _convert_message_to_responses_input(+ message: ChatCompletionMessageParam,+) -> Dict[str, Any]:+ role = message.get("role", "user")+ content = message.get("content", "")++ if isinstance(content, str):+ return {"role": role, "content": content}++ parts: List[Dict[str, Any]] = []+ if isinstance(content, list):+ for part in content:+ if not isinstance(part, dict):+ continue+ if part.get("type") == "text":+ parts.append({"type": "input_text", "text": part.get("text", "")})+ elif part.get("type") == "image_url":+ image_url = part.get("image_url", {})+ parts.append(+ {+ "type": "input_image",+ "image_url": image_url.get("url", ""),+ "detail": image_url.get("detail", "auto"),+ }+ )++ return {"role": role, "content": parts}async def stream_openai_response(@@ -10,49 +39,66 @@ async def stream_openai_response(api_key: str,base_url: str | None,callback: Callable[[str], Awaitable[None]],- model_name: str,+ model: Llm,) -> Completion:start_time = time.time()client = AsyncOpenAI(api_key=api_key, base_url=base_url)- # Base parameters- params = {- "model": model_name,- "messages": messages,- "timeout": 600,- }+ full_response = ""+ if model in OPENAI_CODEX_MODELS:+ input_items = [_convert_message_to_responses_input(msg) for msg in messages]+ reasoning_effort = get_openai_reasoning_effort(model)+ params: Dict[str, Any] = {+ "model": get_openai_api_name(model),+ "input": input_items,+ "stream": True,+ "max_output_tokens": 30000,+ }+ if reasoning_effort:+ params["reasoning"] = {"effort": reasoning_effort}- params["temperature"] = 0- params["stream"] = True+ responses_client = getattr(client, "responses", None)+ if responses_client is None:+ raise Exception(+ "OpenAI SDK is too old for GPT-5.2 Codex. Please upgrade the 'openai' package to a version that supports the Responses API."+ )+ stream = await responses_client.create(**params) # type: ignore+ async for event in stream: # type: ignore+ event_type = getattr(event, "type", None)+ if event_type is None and isinstance(event, dict):+ event_type = event.get("type")+ if event_type == "response.output_text.delta":+ delta = getattr(event, "delta", None)+ if delta is None and isinstance(event, dict):+ delta = event.get("delta", "")+ if not delta:+ continue+ full_response += delta+ await callback(delta)+ else:+ # Base parameters+ params = {+ "model": get_openai_api_name(model),+ "messages": messages,+ "timeout": 600,+ }- # 4.1 series- if model_name in [- "gpt-4.1-2025-04-14",- ]:params["temperature"] = 0params["stream"] = Trueparams["max_tokens"] = 30000- if model_name in [- "gpt-5.2-2025-12-11",- ]:- params["temperature"] = 0- params["stream"] = True- params["max_completion_tokens"] = 30000-- stream = await client.chat.completions.create(**params) # type: ignore- full_response = ""- async for chunk in stream: # type: ignore- assert isinstance(chunk, ChatCompletionChunk)- if (- chunk.choices- and len(chunk.choices) > 0- and chunk.choices[0].delta- and chunk.choices[0].delta.content- ):- content = chunk.choices[0].delta.content or ""- full_response += content- await callback(content)+ stream = await client.chat.completions.create(**params) # type: ignore+ async for chunk in stream: # type: ignore+ assert isinstance(chunk, ChatCompletionChunk)+ if (+ chunk.choices+ and len(chunk.choices) > 0+ and chunk.choices[0].delta+ and chunk.choices[0].delta.content+ ):+ content = chunk.choices[0].delta.content or ""+ full_response += content+ await callback(content)await client.close()diff --git a/backend/poetry.lock b/backend/poetry.lockindex 1107a910..0313d9e5 100644--- a/backend/poetry.lock+++ b/backend/poetry.lock@@ -2558,4 +2558,4 @@ type = ["pytest-mypy"][metadata]lock-version = "2.1"python-versions = "^3.10"-content-hash = "f7c92d17d1545796446109c8f24e360a27f7ec2e26ec5f58f8d1850856252beb"+content-hash = "539fd296366800b71f9dbf90f68f428035a5b9038e785a073747a7728e129630"diff --git a/backend/pyproject.toml b/backend/pyproject.tomlindex 3ec55a67..7c6da75d 100644--- a/backend/pyproject.toml+++ b/backend/pyproject.toml@@ -11,7 +11,7 @@ python = "^3.10"fastapi = "^0.115.6"uvicorn = "^0.25.0"websockets = "^14.1"-openai = "^1.2.4"+openai = "^1.40.0"python-dotenv = "^1.0.0"beautifulsoup4 = "^4.12.2"httpx = "^0.28.1"diff --git a/backend/routes/generate_code.py b/backend/routes/generate_code.pyindex 26c81799..06e3ab0b 100644--- a/backend/routes/generate_code.py+++ b/backend/routes/generate_code.py@@ -416,7 +416,7 @@ class ModelSelectionStage:Llm.GEMINI_3_FLASH_PREVIEW_HIGH,Llm.GEMINI_3_PRO_PREVIEW_HIGH,Llm.CLAUDE_4_5_OPUS_2025_11_01,- Llm.GPT_5_2_2025_12_11,+ Llm.GPT_5_2_CODEX_HIGH,]elif gemini_api_key and anthropic_api_key:models = [@@ -428,14 +428,17 @@ class ModelSelectionStage:]elif gemini_api_key and openai_api_key:models = [- Llm.GPT_5_2_2025_12_11,Llm.GEMINI_3_FLASH_PREVIEW_HIGH,- Llm.GEMINI_3_FLASH_PREVIEW_MINIMAL,Llm.GEMINI_3_PRO_PREVIEW_HIGH,- Llm.GEMINI_3_PRO_PREVIEW_LOW,+ Llm.GPT_5_2_CODEX_HIGH,+ Llm.GPT_5_2_CODEX_MEDIUM,]elif openai_api_key and anthropic_api_key:- models = [Llm.CLAUDE_4_5_OPUS_2025_11_01, Llm.GPT_5_2_2025_12_11]+ models = [+ Llm.CLAUDE_4_5_OPUS_2025_11_01,+ Llm.GPT_5_2_CODEX_HIGH,+ Llm.GPT_5_2_CODEX_MEDIUM,+ ]elif gemini_api_key:models = [Llm.GEMINI_3_FLASH_PREVIEW_HIGH,@@ -446,7 +449,7 @@ class ModelSelectionStage:elif anthropic_api_key:models = [Llm.CLAUDE_4_5_OPUS_2025_11_01]elif openai_api_key:- models = [Llm.GPT_5_2_2025_12_11]+ models = [Llm.GPT_5_2_CODEX_HIGH, Llm.GPT_5_2_CODEX_MEDIUM]else:raise Exception("No OpenAI or Anthropic key")@@ -655,7 +658,7 @@ class ParallelGenerationStage:tasks.append(self._stream_openai_with_error_handling(prompt_messages,- model_name=model.value,+ model=model,index=index,))@@ -700,7 +703,7 @@ class ParallelGenerationStage:async def _stream_openai_with_error_handling(self,prompt_messages: List[ChatCompletionMessageParam],- model_name: str,+ model: Llm,index: int,) -> Completion:"""Wrap OpenAI streaming with specific error handling"""@@ -711,7 +714,7 @@ class ParallelGenerationStage:api_key=self.openai_api_key,base_url=self.openai_base_url,callback=lambda x: self._process_chunk(x, index),- model_name=model_name,+ model=model,)except openai.AuthenticationError as e:print(f"[VARIANT {index + 1}] OpenAI Authentication failed", e)diff --git a/backend/tests/test_model_selection.py b/backend/tests/test_model_selection.pyindex ef144150..3c218287 100644--- a/backend/tests/test_model_selection.py+++ b/backend/tests/test_model_selection.py@@ -14,7 +14,7 @@ class TestModelSelectionAllKeys:@pytest.mark.asyncioasync def test_gemini_anthropic_create(self):- """All keys: Claude 4.5 Opus, GPT 5.2, Gemini 3 Flash (high), Gemini 3 Pro (high)"""+ """All keys: Claude 4.5 Opus, GPT 5.2 Codex (high), Gemini 3 Flash (high), Gemini 3 Pro (high)"""models = await self.model_selector.select_models(generation_type="create",input_mode="text",@@ -27,7 +27,7 @@ class TestModelSelectionAllKeys:Llm.GEMINI_3_FLASH_PREVIEW_HIGH,Llm.GEMINI_3_PRO_PREVIEW_HIGH,Llm.CLAUDE_4_5_OPUS_2025_11_01,- Llm.GPT_5_2_2025_12_11,+ Llm.GPT_5_2_CODEX_HIGH,]assert models == expected@@ -46,7 +46,7 @@ class TestModelSelectionAllKeys:Llm.GEMINI_3_FLASH_PREVIEW_HIGH,Llm.GEMINI_3_PRO_PREVIEW_HIGH,Llm.CLAUDE_4_5_OPUS_2025_11_01,- Llm.GPT_5_2_2025_12_11,+ Llm.GPT_5_2_CODEX_HIGH,]assert models == expected@@ -61,7 +61,7 @@ class TestModelSelectionOpenAIAnthropic:@pytest.mark.asyncioasync def test_openai_anthropic(self):- """OpenAI + Anthropic: Claude 4.5 Opus, GPT 5.2, cycling"""+ """OpenAI + Anthropic: Claude 4.5 Opus, GPT 5.2 Codex (high/medium), cycling"""models = await self.model_selector.select_models(generation_type="create",input_mode="text",@@ -72,9 +72,9 @@ class TestModelSelectionOpenAIAnthropic:expected = [Llm.CLAUDE_4_5_OPUS_2025_11_01,- Llm.GPT_5_2_2025_12_11,+ Llm.GPT_5_2_CODEX_HIGH,+ Llm.GPT_5_2_CODEX_MEDIUM,Llm.CLAUDE_4_5_OPUS_2025_11_01,- Llm.GPT_5_2_2025_12_11,]assert models == expected@@ -117,7 +117,7 @@ class TestModelSelectionOpenAIOnly:@pytest.mark.asyncioasync def test_openai_only(self):- """OpenAI only: GPT 5.2 only"""+ """OpenAI only: GPT 5.2 Codex (high/medium) only"""models = await self.model_selector.select_models(generation_type="create",input_mode="text",@@ -127,10 +127,10 @@ class TestModelSelectionOpenAIOnly:)expected = [- Llm.GPT_5_2_2025_12_11,- Llm.GPT_5_2_2025_12_11,- Llm.GPT_5_2_2025_12_11,- Llm.GPT_5_2_2025_12_11,+ Llm.GPT_5_2_CODEX_HIGH,+ Llm.GPT_5_2_CODEX_MEDIUM,+ Llm.GPT_5_2_CODEX_HIGH,+ Llm.GPT_5_2_CODEX_MEDIUM,]assert models == expecteddiff --git a/frontend/src/lib/models.ts b/frontend/src/lib/models.tsindex b496e8df..5e0daee2 100644--- a/frontend/src/lib/models.ts+++ b/frontend/src/lib/models.ts@@ -1,7 +1,8 @@// Keep in sync with backend (llm.py)// Order here matches dropdown orderexport enum CodeGenerationModel {- GPT_5_2_2025_12_11 = "gpt-5.2-2025-12-11",+ GPT_5_2_CODEX_MEDIUM = "gpt-5.2-codex (medium thinking)",+ GPT_5_2_CODEX_HIGH = "gpt-5.2-codex (high thinking)",CLAUDE_4_5_OPUS_2025_11_01 = "claude-opus-4-5-20251101",CLAUDE_4_5_SONNET_2025_09_29 = "claude-sonnet-4-5-20250929",GEMINI_3_FLASH_PREVIEW_HIGH = "gemini-3-flash-preview (high thinking)",@@ -14,7 +15,14 @@ export enum CodeGenerationModel {export const CODE_GENERATION_MODEL_DESCRIPTIONS: {[key in CodeGenerationModel]: { name: string; inBeta: boolean };} = {- "gpt-5.2-2025-12-11": { name: "GPT 5.2", inBeta: false },+ "gpt-5.2-codex (medium thinking)": {+ name: "GPT 5.2 Codex (medium)",+ inBeta: true,+ },+ "gpt-5.2-codex (high thinking)": {+ name: "GPT 5.2 Codex (high)",+ inBeta: true,+ },"claude-opus-4-5-20251101": { name: "Claude Opus 4.5", inBeta: false },"claude-sonnet-4-5-20250929": { name: "Claude Sonnet 4.5", inBeta: true },"gemini-3-flash-preview (high thinking)": {