agenthub-python 0.4.10__tar.gz → 0.4.11__tar.gz

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agenthub-python
-Version: 0.4.10
+Version: 0.4.11
 Summary: AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents.
 Keywords: agent,llm,gemini,claude,gpt
 Author: PrismShadow

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/agenthub/auto_client.py

@@ -84,10 +84,12 @@ class AutoLLMClient(LLMClient):
             from .claude5 import Claude5Client
 
             return Claude5Client
-        elif "gpt-5.4" in client_type or "gpt-5.5" in client_type or "gpt-5.6" in client_type:  # e.g., gpt-5.6
-            from .gpt5_6 import GPT5_6Client
+        elif (
+            "gpt-5.4" in client_type or "gpt-5.5" in client_type or "gpt-5.6" in client_type or "gpt-6" in client_type
+        ):  # e.g., gpt-6-astra
+            from .gpt6 import GPT6Client
 
-            return GPT5_6Client
+            return GPT6Client
         elif "glm-5" in client_type:  # the whole GLM series shares the unified client
             from .glm5_3 import GLM5_3Client
 
@@ -144,7 +146,7 @@ class AutoLLMClient(LLMClient):
             raise ValueError(
                 f"{client_type} is not supported. "
                 "Supported client types: minimax-m3, gemini-3.8, gemini-3.7, gemini-3.6, gemini-3, "
-                "claude-5, claude-4-8, claude-4-7, claude-4-6, gpt-5.6, gpt-5.5, gpt-5.4, "
+                "claude-5, claude-4-8, claude-4-7, claude-4-6, gpt-6, gpt-5.6, gpt-5.5, gpt-5.4, "
                 "glm-5.3, glm-5.2, glm-5.1, kimi-k3, kimi-k2.6, kimi-k2.5, deepseek-v4, "
                 "openai-chat-vllm-adapter, openai-embedding, ant-messages, openai-responses, openai-chat."
             )

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/agenthub/deepseek_v4/client.py

@@ -14,6 +14,7 @@
 
 import json
 import os
+import re
 from typing import Any, AsyncIterator
 
 from openai import AsyncOpenAI
@@ -36,6 +37,13 @@ from ..types import (
 from ..utils import is_debug_enabled
 
 
+# The DeepSeek ids that read no image: the current V4 Flash and V4 Pro, bare or with a dated
+# snapshot suffix (deepseek-v4-flash-0731). Every other id forwards its images. Matched against
+# the bare id — the part after the last "/", lowercased — so a gateway prefix (deepseek/,
+# deepseek-ai/) and the spelling a platform uses do not change the verdict.
+_TEXT_ONLY_MODELS = re.compile(r"deepseek-v4-(flash|pro)(-\d{4})?")
+
+
 class DeepSeekV4Client(LLMClient):
     """DeepSeek V4-specific LLM client implementation using the OpenAI-compatible Responses API."""
 
@@ -142,10 +150,10 @@ class DeepSeekV4Client(LLMClient):
         Returns:
             List of input items for the Responses API
         """
-        # only a vision model reads image parts; every other DeepSeek model answers from a
-        # placeholder instead of failing (llmsdk_docs/deepseek_v4/docs/responses-api.md), so an
-        # image is refused here rather than silently dropped
-        supports_image = "vision" in self._model.lower()
+        # a text-only model answers from a placeholder instead of failing
+        # (llmsdk_docs/deepseek_v4/docs/responses-api.md), so an image is refused here rather
+        # than silently dropped
+        supports_image = not _TEXT_ONLY_MODELS.fullmatch(self._model.lower().rsplit("/", 1)[-1])
         input_list: list[ResponseInputParam] = []
 
         for msg in messages:

{agenthub_python-0.4.10/agenthub/gpt5_6 → agenthub_python-0.4.11/agenthub/gpt6}/__init__.py

@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from .client import GPT5_6Client
+from .client import GPT6Client
 
 
-__all__ = ["GPT5_6Client"]
+__all__ = ["GPT6Client"]

{agenthub_python-0.4.10/agenthub/gpt5_6 → agenthub_python-0.4.11/agenthub/gpt6}/client.py

@@ -36,8 +36,8 @@ from ..types import (
 from ..utils import is_debug_enabled, openai_image_detail
 
 
-class GPT5_6Client(LLMClient):
-    """GPT-5.6-specific LLM client implementation (also serves GPT-5.4 and GPT-5.5)."""
+class GPT6Client(LLMClient):
+    """GPT-6-specific LLM client implementation (also serves GPT-5.6, GPT-5.5 and GPT-5.4)."""
 
     def __init__(
         self,
@@ -46,7 +46,7 @@ class GPT5_6Client(LLMClient):
         base_url: str | None = None,
         default_headers: dict[str, str] | None = None,
     ):
-        """Initialize GPT-5.6 client with model and API key."""
+        """Initialize GPT-6 client with model and API key."""
         self._model = model
         api_key = api_key or os.getenv("OPENAI_API_KEY")
         base_url = base_url or os.getenv("OPENAI_BASE_URL")
@@ -55,6 +55,13 @@ class GPT5_6Client(LLMClient):
 
     def _convert_thinking_level_to_effort(self, thinking_level: ThinkingLevel) -> str:
         """Convert ThinkingLevel enum to OpenAI's reasoning effort."""
+        if thinking_level == ThinkingLevel.NONE and "gpt-6" in self._model:
+            # GPT-6 rejects both "none" and "minimal" with a 400 (verified live 2026-09-09:
+            # "Unsupported value: 'none' is not supported with the 'gpt-6-astra' model.
+            # Supported values are: 'low', 'medium', 'high', 'xhigh', and 'max'."), so NONE
+            # degrades to the lowest effort the generation accepts.
+            return "low"
+
         mapping = {
             ThinkingLevel.NONE: "none",
             ThinkingLevel.LOW: "low",
@@ -101,7 +108,7 @@ class GPT5_6Client(LLMClient):
 
         if config.get("temperature") is not None and config["temperature"] != 1.0:
             raise UnsupportedParameterError(
-                self.__class__.__name__, "temperature", "GPT-5.6 does not support setting temperature."
+                self.__class__.__name__, "temperature", "GPT-6 does not support setting temperature."
             )
 
         if config.get("thinking_level") is not None:
@@ -124,7 +131,7 @@ class GPT5_6Client(LLMClient):
 
         if config.get("prompt_caching") is not None and config["prompt_caching"] != PromptCaching.ENABLE:
             raise UnsupportedParameterError(
-                self.__class__.__name__, "prompt_caching", "prompt_caching must be ENABLE for GPT-5.6."
+                self.__class__.__name__, "prompt_caching", "prompt_caching must be ENABLE for GPT-6."
             )
 
         return openai_config
@@ -306,7 +313,12 @@ class GPT5_6Client(LLMClient):
                 # the completed item carries the canonical wire fields to send back on the
                 # next turn (identical to the response.completed copy, but adjacent to the
                 # thinking deltas so the fidelity lands on the item that carried the text);
-                # record the channel plus the fields the server demands back
+                # record the channel plus the fields the server demands back. This event is the
+                # only source of encrypted_content, because the streaming-events reference says
+                # of response.output_item.added: "For reasoning items, encrypted_content may be
+                # incomplete while the item is in progress. Use the reasoning item from the
+                # corresponding response.output_item.done event when passing it as input to a
+                # subsequent request."
                 event_type = "delta"
                 fidelity = {}
                 if getattr(model_output.item, "summary", None):

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/agenthub/openai_chat/client.py

@@ -173,7 +173,7 @@ class OpenaiChatClient(LLMClient):
                     if "tool_call_id" not in item:
                         raise ValueError("tool_call_id is required for tool result.")
 
-                    content = [{"type": "text", "text": item["text"]}]
+                    image_parts = []
 
                     if "images" in item and item["images"]:
                         for image_url in item["images"]:
@@ -182,7 +182,12 @@ class OpenaiChatClient(LLMClient):
                                 # siliconflow does not support image_url in tool result
                                 content_parts.append(part)
                             else:
-                                content.append(part)
+                                image_parts.append(part)
+
+                    # a plain string is the form every OpenAI-compatible server accepts for a text
+                    # result; the content-part list is reserved for results carrying images, which
+                    # only servers with multimodal tool messages take
+                    content = [{"type": "text", "text": item["text"]}, *image_parts] if image_parts else item["text"]
 
                     # Tool results are sent as separate messages
                     openai_messages.append(

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/agenthub/openai_responses/client.py

@@ -55,6 +55,12 @@ class OpenaiResponsesClient(LLMClient):
 
     def _convert_thinking_level_to_effort(self, thinking_level: ThinkingLevel) -> str:
         """Convert ThinkingLevel enum to the Responses API reasoning effort."""
+        if thinking_level == ThinkingLevel.NONE and "gpt-6" in self._model:
+            # a gateway serving GPT-6 forwards the effort to OpenAI, which rejects "none" and
+            # "minimal" with a 400 (verified live 2026-09-09 against api.openai.com), so NONE
+            # degrades to the lowest effort the generation accepts.
+            return "low"
+
         mapping = {
             ThinkingLevel.NONE: "none",
             ThinkingLevel.LOW: "low",
@@ -202,13 +208,20 @@ class OpenaiResponsesClient(LLMClient):
                         raise ValueError("tool_call_id is required for tool result.")
 
                     # NOTE: tool results are input items
-                    tool_result = [{"type": "input_text", "text": item["text"]}]
+                    image_parts = []
                     if "images" in item:
                         for image_url in item["images"]:
-                            tool_result.append(self._convert_image_url(image_url))
+                            image_parts.append(self._convert_image_url(image_url))
+
+                    # a plain string is the form every OpenAI-compatible server accepts for a text
+                    # result; the content-part list is reserved for results carrying images, which
+                    # only servers with multimodal tool messages take
+                    output = (
+                        [{"type": "input_text", "text": item["text"]}, *image_parts] if image_parts else item["text"]
+                    )
 
                     input_list.append(
-                        {"type": "function_call_output", "call_id": item["tool_call_id"], "output": tool_result}
+                        {"type": "function_call_output", "call_id": item["tool_call_id"], "output": output}
                     )
                 else:
                     raise ValueError(f"Unknown item: {item}")

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/agenthub/registry.py

@@ -212,11 +212,24 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
         "context_window": 1000000,
         "pricing": _usd(3.0, 15.0, cached=0.3),
     },
+    {
+        # official list price per 1M tokens: $1 cached input, $10 uncached input, $12.5 cache
+        # writes, $50 output, all doubled above 272K input tokens except output, which is 1.5x.
+        # The prompt bucket carries the cache-write rate: the usage buckets do not separate
+        # cache-written input from plain input.
+        "model": "gpt-6-astra",
+        "base_url": _OPENAI,
+        "client": "gpt-6",
+        "input_modalities": ["Text", "Image"],
+        "output_modalities": ["Text"],
+        "context_window": 1050000,
+        "pricing": _usd(12.5, 50.0, cached=1.0),
+    },
     {
         # official standard-tier list price; the bare gpt-5.6 alias also routes here
         "model": "gpt-5.6-sol",
         "base_url": _OPENAI,
-        "client": "gpt-5.6",
+        "client": "gpt-6",
         "input_modalities": ["Text", "Image"],
         "output_modalities": ["Text"],
         "context_window": 1050000,
@@ -225,7 +238,7 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
     {
         "model": "gpt-5.6-terra",
         "base_url": _OPENAI,
-        "client": "gpt-5.6",
+        "client": "gpt-6",
         "input_modalities": ["Text", "Image"],
         "output_modalities": ["Text"],
         "context_window": 1050000,
@@ -234,7 +247,7 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
     {
         "model": "gpt-5.6-luna",
         "base_url": _OPENAI,
-        "client": "gpt-5.6",
+        "client": "gpt-6",
         "input_modalities": ["Text", "Image"],
         "output_modalities": ["Text"],
         "context_window": 1050000,
@@ -323,6 +336,20 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
         "context_window": 262144,
         "pricing": _cny(6.5, 27.0, cached=1.1),
     },
+    {
+        "model": "deepseek-v4.1-flash",
+        "base_url": _DEEPSEEK,
+        "client": "deepseek-v4",
+        # announced by DeepSeek for release after 2026-09-10 and not yet served on
+        # 2026-09-09; multimodal per the announcement
+        "input_modalities": ["Text", "Image"],
+        "output_modalities": ["Text"],
+        # assumed equal to deepseek-v4-flash until the official model page lists it
+        "context_window": 1000000,
+        # priced as the V4 Flash series: official off-peak list price effective 2026-09-10,
+        # and peak-hour rates (Beijing 9:00-12:00, 14:00-18:00) are double
+        "pricing": _cny(1.0, 4.0, cached=0.02),
+    },
     {
         "model": "deepseek-v4-flash",
         "base_url": _DEEPSEEK,
@@ -330,9 +357,10 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
         "input_modalities": ["Text"],
         "output_modalities": ["Text"],
         "context_window": 1000000,
-        # official off-peak list price (verified 2026-08-18); peak-hour rates
-        # (Beijing 9:00-12:00, 14:00-18:00) are double
-        "pricing": _cny(1.5, 4.5, cached=0.05),
+        # official off-peak list price effective 2026-09-10 (verified 2026-09-08 against
+        # the official adjustment); peak-hour rates (Beijing 9:00-12:00, 14:00-18:00) are
+        # double
+        "pricing": _cny(1.0, 4.0, cached=0.02),
     },
     {
         "model": "deepseek-v4-flash-vision-exp",
@@ -341,9 +369,10 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
         "input_modalities": ["Text", "Image"],
         "output_modalities": ["Text"],
         "context_window": 1000000,
-        # priced as deepseek-v4-flash; official off-peak list price (verified 2026-08-21), and
-        # peak-hour rates (Beijing 9:00-12:00, 14:00-18:00) are double
-        "pricing": _cny(1.5, 4.5, cached=0.05),
+        # priced as deepseek-v4-flash; official off-peak list price effective 2026-09-10
+        # (verified 2026-09-08 against the official adjustment), and peak-hour rates
+        # (Beijing 9:00-12:00, 14:00-18:00) are double
+        "pricing": _cny(1.0, 4.0, cached=0.02),
     },
     {
         "model": "deepseek-v4-pro",
@@ -465,6 +494,17 @@ _SUPPORTED_MODELS: list[SupportedModel] = [
         "context_window": 1000000,
         "pricing": _usd(0.0, 0.0),
     },
+    {
+        # models API 2026-09-09, default OpenAI endpoint, no discount: $10 input, $50 output,
+        # $1 cache read, $12.5 cache write per 1M tokens
+        "model": "openai/gpt-6-astra",
+        "base_url": _OPENROUTER,
+        "client": "openai-responses",
+        "input_modalities": ["Text", "Image"],
+        "output_modalities": ["Text"],
+        "context_window": 1050000,
+        "pricing": _usd(12.5, 50.0, cached=1.0),
+    },
     {
         "model": "openai/gpt-5.6-sol",
         "base_url": _OPENROUTER,

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agenthub-python"
-version = "0.4.10"
+version = "0.4.11"
 description = "AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents."
 keywords = [
     "agent",

{agenthub_python-0.4.10 → agenthub_python-0.4.11}/pyproject.toml.orig

@@ -1,6 +1,6 @@
 [project]
 name = "agenthub-python"
-version = "0.4.10"
+version = "0.4.11"
 description = "AgentHub is the LLM API Hub for the Agent era, built for high-precision autonomous agents."
 keywords = ["agent", "llm", "gemini", "claude", "gpt"]
 readme = "README.md"