unique_toolkit 2026.40.0rc2__tar.gz → 2026.42.0.dev0__tar.gz

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: unique-toolkit
-Version: 2026.40.0rc2
+Version: 2026.42.0.dev0
 Summary: 
 Author: Cedric Klinkert, Martin Fadler, Sadique Sheik, Fabian Schläpfer, Pascal Hauri, Ejiroghene Oki-Peter, Rami Azouz, Ricardo Visini, Sanzio Monti
 Author-email: Cedric Klinkert <cedric.klinkert@unique.ch>, Martin Fadler <martin.fadler@unique.ch>, Sadique Sheik <sadique@unique.ch>, Fabian Schläpfer <fabian@unique.ch>, Pascal Hauri <pascal@unique.ch>, Ejiroghene Oki-Peter <oki-peter.ejiroghene@unique.ai>, Rami Azouz <rami.ext@unique.ch>, Ricardo Visini <ricardo.visini@unique.ch>, Sanzio Monti <sanzio.monti@unique.ai>
@@ -16,7 +16,7 @@ Requires-Dist: sseclient>=0.0.27,<0.0.28
 Requires-Dist: openai>=1.105.0,<3
 Requires-Dist: platformdirs>=4.0.0,<5
 Requires-Dist: pillow>=12.1.1
-Requires-Dist: unique-sdk>=2026.40.0rc2
+Requires-Dist: unique-sdk>=2026.42.0.dev0,<2026.42.0rc0
 Requires-Dist: jambo>=0.1.2,<0.2
 Requires-Dist: jinja2>=3.1.6,<4
 Requires-Dist: docxtpl>=0.20.1,<0.21

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "unique_toolkit"
-version = "2026.40.0rc2"
+version = "2026.42.0.dev0"
 description = ""
 readme = "README.md"
 requires-python = ">=3.12"
@@ -28,7 +28,7 @@ dependencies = [
     "openai>=1.105.0,<3",
     "platformdirs>=4.0.0,<5",
     "pillow>=12.1.1",
-    "unique-sdk>=2026.40.0rc2",
+    "unique-sdk>=2026.42.0.dev0,<2026.42.0rc0",
     "jambo>=0.1.2,<0.2",
     "jinja2>=3.1.6,<4",
     "docxtpl>=0.20.1,<0.21",

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/unique_toolkit/agentic/loop_runner/__init__.py

@@ -21,6 +21,7 @@ from unique_toolkit.agentic.loop_runner.runners import (
     BasicLoopIterationRunner,
     BasicLoopIterationRunnerConfig,
     MistralLoopIterationRunner,
+    PromptForcedToolLoopIterationRunner,
     QwenLoopIterationRunner,
     ResponsesBasicLoopIterationRunner,
 )
@@ -34,6 +35,7 @@ __all__ = [
     "BasicLoopIterationRunnerConfig",
     "BasicLoopIterationRunner",
     "MistralLoopIterationRunner",
+    "PromptForcedToolLoopIterationRunner",
     "QwenLoopIterationRunner",
     "handle_forced_tools_iteration",
     "handle_last_iteration",

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/unique_toolkit/agentic/loop_runner/runners/__init__.py

@@ -6,6 +6,10 @@ from unique_toolkit.agentic.loop_runner.runners.basic import (
 from unique_toolkit.agentic.loop_runner.runners.mistral import (
     MistralLoopIterationRunner,
 )
+from unique_toolkit.agentic.loop_runner.runners.prompt_forced_tool import (
+    PROMPT_FORCED_TOOL_CALL_INSTRUCTION,
+    PromptForcedToolLoopIterationRunner,
+)
 from unique_toolkit.agentic.loop_runner.runners.qwen import (
     QWEN_FORCED_TOOL_CALL_INSTRUCTION,
     QWEN_LAST_ITERATION_INSTRUCTION,
@@ -18,6 +22,8 @@ __all__ = [
     "BasicLoopIterationRunner",
     "ResponsesBasicLoopIterationRunner",
     "MistralLoopIterationRunner",
+    "PROMPT_FORCED_TOOL_CALL_INSTRUCTION",
+    "PromptForcedToolLoopIterationRunner",
     "QwenLoopIterationRunner",
     "QWEN_FORCED_TOOL_CALL_INSTRUCTION",
     "QWEN_LAST_ITERATION_INSTRUCTION",

unique_toolkit-2026.42.0.dev0/unique_toolkit/agentic/loop_runner/runners/prompt_forced_tool/__init__.py

@@ -0,0 +1,9 @@
+from unique_toolkit.agentic.loop_runner.runners.prompt_forced_tool.runner import (
+    PROMPT_FORCED_TOOL_CALL_INSTRUCTION,
+    PromptForcedToolLoopIterationRunner,
+)
+
+__all__ = [
+    "PROMPT_FORCED_TOOL_CALL_INSTRUCTION",
+    "PromptForcedToolLoopIterationRunner",
+]

unique_toolkit-2026.42.0.dev0/unique_toolkit/agentic/loop_runner/runners/prompt_forced_tool/helpers.py

@@ -0,0 +1,32 @@
+"""Message helpers for models that reject a forced tool_choice."""
+
+from unique_toolkit.language_model.schemas import (
+    LanguageModelMessageRole,
+    LanguageModelMessages,
+)
+
+
+def append_prompt_forced_tool_instruction(
+    *,
+    messages: LanguageModelMessages,
+    instruction: str,
+) -> LanguageModelMessages:
+    """Append a must-call instruction to the last user message.
+
+    Args:
+        messages (LanguageModelMessages): Conversation copied for this tool.
+        instruction (str): Instruction that names the tool to call.
+
+    Returns:
+        LanguageModelMessages: A new message list with the instruction appended
+        when the last user message has string content.
+    """
+    messages_list = list(messages)
+    for i in range(len(messages_list) - 1, -1, -1):
+        msg = messages_list[i]
+        if msg.role == LanguageModelMessageRole.USER and isinstance(msg.content, str):
+            messages_list[i] = msg.model_copy(
+                update={"content": msg.content + "\n" + instruction}
+            )
+            break
+    return LanguageModelMessages(root=messages_list)

unique_toolkit-2026.42.0.dev0/unique_toolkit/agentic/loop_runner/runners/prompt_forced_tool/runner.py

@@ -0,0 +1,53 @@
+"""Loop runner for models that reject forced tool_choice values."""
+
+from typing import Unpack
+
+from unique_toolkit.agentic.loop_runner._iteration_handler_utils import (
+    run_forced_tools_iteration,
+)
+from unique_toolkit.agentic.loop_runner.base import _LoopIterationRunnerKwargs
+from unique_toolkit.agentic.loop_runner.runners.basic import (
+    BasicLoopIterationRunner,
+)
+from unique_toolkit.agentic.loop_runner.runners.prompt_forced_tool.helpers import (
+    append_prompt_forced_tool_instruction,
+)
+from unique_toolkit.chat.functions import LanguageModelStreamResponse
+
+PROMPT_FORCED_TOOL_CALL_INSTRUCTION = (
+    "You must call the tool {TOOL_NAME}. Call it instead of replying with text."
+)
+
+
+class PromptForcedToolLoopIterationRunner(BasicLoopIterationRunner):
+    """Runner for models that reject named and ``any`` tool choices.
+
+    Claude Opus 5.5 and Fable 5.1 return HTTP 400 for those ``tool_choice``
+    types. This runner keeps one request per forced tool, limits ``tools`` to
+    that tool, sends ``tool_choice="auto"``, and asks for the call in the prompt.
+    """
+
+    async def _handle_forced_tools(
+        self,
+        **kwargs: Unpack[_LoopIterationRunnerKwargs],
+    ) -> LanguageModelStreamResponse:
+        original_messages = kwargs["messages"].model_copy(deep=True)
+
+        def _prepare(
+            func_name: str | None,
+            per_choice_kwargs: _LoopIterationRunnerKwargs,
+        ) -> _LoopIterationRunnerKwargs:
+            instruction = PROMPT_FORCED_TOOL_CALL_INSTRUCTION.format(
+                TOOL_NAME=func_name or ""
+            )
+            per_choice_kwargs["messages"] = append_prompt_forced_tool_instruction(
+                messages=original_messages,
+                instruction=instruction,
+            )
+            return per_choice_kwargs
+
+        return await run_forced_tools_iteration(
+            loop_runner_kwargs=kwargs,
+            prepare_loop_runner_kwargs=_prepare,
+            tool_choice_override="auto",
+        )

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/unique_toolkit/agentic_table/service.py

@@ -653,6 +653,8 @@ class AgenticTableService:
         row_orders: list[int],
         status: RowVerificationStatus,
         locked: bool | None = None,
+        *,
+        emit_library_sheet_verified_event: bool | None = None,
     ):
         """Update the verification status of multiple rows at once.
 
@@ -660,24 +662,27 @@ class AgenticTableService:
             row_orders: The row indexes to update.
             status: The verification status to set (``NEEDS_REVIEW``, etc.; matches public API).
             locked: When set, forwarded as ``locked`` on ``POST .../rows/bulk-update-status``.
+            emit_library_sheet_verified_event: When set, forwarded as
+                ``emitLibrarySheetVerifiedEvent``. ``False`` marks the rows without
+                starting a ``LIBRARY_SHEET_ROW_VERIFIED`` run on library sheets; use it
+                when the caller has already ingested the rows. ``None`` omits the field
+                and keeps the server default (``True``). Servers that predate the field
+                reject it with a 400, so only set it against a backend that supports it.
         """
-        if locked is None:
-            await AgenticTable.bulk_update_status(
-                user_id=self._user_id,
-                company_id=self._company_id,
-                tableId=self.table_id,
-                rowOrders=row_orders,
-                status=status,
-            )
-        else:
-            await AgenticTable.bulk_update_status(
-                user_id=self._user_id,
-                company_id=self._company_id,
-                tableId=self.table_id,
-                rowOrders=row_orders,
-                status=status,
-                locked=locked,
-            )
+        params: AgenticTable.BulkUpdateStatus = {
+            "tableId": self.table_id,
+            "rowOrders": row_orders,
+            "status": status,
+        }
+        if locked is not None:
+            params["locked"] = locked
+        if emit_library_sheet_verified_event is not None:
+            params["emitLibrarySheetVerifiedEvent"] = emit_library_sheet_verified_event
+        await AgenticTable.bulk_update_status(
+            user_id=self._user_id,
+            company_id=self._company_id,
+            **params,
+        )
 
     async def import_questions_and_sources(
         self,

{unique_toolkit-2026.40.0rc2 → unique_toolkit-2026.42.0.dev0}/unique_toolkit/language_model/infos.py

@@ -66,6 +66,8 @@ class LanguageModelName(StrEnum):
     AZURE_GPT_56_TERRA_2026_0709 = "AZURE_GPT_56_TERRA_2026_0709"
     AZURE_GPT_56_LUNA_2026_0709 = "AZURE_GPT_56_LUNA_2026_0709"
     AZURE_GPT_6_ASTRA_2026_0903 = "AZURE_GPT_6_ASTRA_2026_0903"
+    AZURE_GPT_6_LUNA_2026_0922 = "AZURE_GPT_6_LUNA_2026_0922"
+    AZURE_GPT_6_SOL_2026_0922 = "AZURE_GPT_6_SOL_2026_0922"
     AZURE_GPT_4o_2024_0513 = "AZURE_GPT_4o_2024_0513"
     AZURE_GPT_4o_2024_0806 = "AZURE_GPT_4o_2024_0806"
     AZURE_GPT_4o_2024_1120 = "AZURE_GPT_4o_2024_1120"
@@ -96,6 +98,7 @@ class LanguageModelName(StrEnum):
     ANTHROPIC_CLAUDE_OPUS_4_7 = "litellm:anthropic-claude-opus-4-7"
     ANTHROPIC_CLAUDE_OPUS_4_8 = "litellm:anthropic-claude-opus-4-8"
     ANTHROPIC_CLAUDE_OPUS_5 = "litellm:anthropic-claude-opus-5"
+    ANTHROPIC_CLAUDE_OPUS_5_5 = "litellm:anthropic-claude-opus-5-5"
     ANTHROPIC_CLAUDE_FABLE_5 = "litellm:anthropic-claude-fable-5"
     GEMINI_2_0_FLASH = "litellm:gemini-2-0-flash"
     GEMINI_2_5_FLASH = "litellm:gemini-2-5-flash"
@@ -107,12 +110,14 @@ class LanguageModelName(StrEnum):
     GEMINI_2_5_PRO_PREVIEW_0605 = "litellm:gemini-2-5-pro-preview-06-05"
     GEMINI_3_1_PRO_PREVIEW = "litellm:gemini-3-1-pro-preview"
     GEMINI_3_6_FLASH = "litellm:gemini-3-6-flash"
+    GEMINI_3_8_FLASH = "litellm:gemini-3-8-flash"
     GEMINI_3_FLASH_PREVIEW = "litellm:gemini-3-flash-preview"
     GEMINI_3_PRO_PREVIEW = "litellm:gemini-3-pro-preview"
     GROK_4_1_FAST_NON_REASONING = "litellm:grok-4-1-fast-non-reasoning"
     GROK_4_1_FAST_REASONING = "litellm:grok-4-1-fast-reasoning"
     GROK_4_5 = "litellm:grok-4-5"
     GROK_4_6 = "litellm:grok-4-6"
+    GROK_4_7 = "litellm:grok-4-7"
     LITELLM_OPENAI_GPT_5 = "litellm:openai-gpt-5"
     LITELLM_OPENAI_GPT_5_MINI = "litellm:openai-gpt-5-mini"
     LITELLM_OPENAI_GPT_5_NANO = "litellm:openai-gpt-5-nano"
@@ -130,6 +135,8 @@ class LanguageModelName(StrEnum):
     LITELLM_OPENAI_GPT_56_TERRA = "litellm:openai-gpt-5-6-terra"
     LITELLM_OPENAI_GPT_56_LUNA = "litellm:openai-gpt-5-6-luna"
     LITELLM_OPENAI_GPT_6_ASTRA = "litellm:openai-gpt-6-astra"
+    LITELLM_OPENAI_GPT_6_LUNA = "litellm:openai-gpt-6-luna"
+    LITELLM_OPENAI_GPT_6_SOL = "litellm:openai-gpt-6-sol"
     LITELLM_OPENAI_O1 = "litellm:openai-o1"
     LITELLM_OPENAI_O3 = "litellm:openai-o3"
     LITELLM_OPENAI_O3_DEEP_RESEARCH = "litellm:openai-o3-deep-research"
@@ -141,6 +148,7 @@ class LanguageModelName(StrEnum):
     LITELLM_DEEPSEEK_R1 = "litellm:deepseek-r1"
     LITELLM_DEEPSEEK_V3 = "litellm:deepseek-v3-1"
     LITELLM_DEEPSEEK_V4_PRO = "litellm:deepseek-v4-pro"
+    LITELLM_DEEPSEEK_V4_1_FLASH = "litellm:deepseek-v4-1-flash"
     LITELLM_GLM_5_1 = "litellm:glm-5.1"
     LITELLM_GLM_5_2 = "litellm:glm-5.2"
     LITELLM_PHOENIQS_DEEPSEEK_V4_FLASH = "litellm:phoeniqs-deepseek-v4-flash"
@@ -156,6 +164,7 @@ class LanguageModelName(StrEnum):
     VERTEX_CLAUDE_OPUS_4_7 = "litellm:vertex-claude-opus-4-7"
     VERTEX_CLAUDE_OPUS_4_8 = "litellm:vertex-claude-opus-4-8"
     VERTEX_CLAUDE_OPUS_5 = "litellm:vertex-claude-opus-5"
+    VERTEX_CLAUDE_OPUS_5_5 = "litellm:vertex-claude-opus-5-5"
     VERTEX_CLAUDE_FABLE_5 = "litellm:vertex-claude-fable-5"
 
 
@@ -224,6 +233,8 @@ def get_encoder_name(model_name: LanguageModelName) -> EncoderName:
             | LMN.AZURE_GPT_56_TERRA_2026_0709
             | LMN.AZURE_GPT_56_LUNA_2026_0709
             | LMN.AZURE_GPT_6_ASTRA_2026_0903
+            | LMN.AZURE_GPT_6_LUNA_2026_0922
+            | LMN.AZURE_GPT_6_SOL_2026_0922
             | LMN.AZURE_MODEL_ROUTER_2025_1118
             | LMN.LITELLM_OPENAI_GPT_5
             | LMN.LITELLM_OPENAI_GPT_5_MINI
@@ -242,6 +253,8 @@ def get_encoder_name(model_name: LanguageModelName) -> EncoderName:
             | LMN.LITELLM_OPENAI_GPT_56_TERRA
             | LMN.LITELLM_OPENAI_GPT_56_LUNA
             | LMN.LITELLM_OPENAI_GPT_6_ASTRA
+            | LMN.LITELLM_OPENAI_GPT_6_LUNA
+            | LMN.LITELLM_OPENAI_GPT_6_SOL
             | LMN.LITELLM_OPENAI_O1
             | LMN.LITELLM_OPENAI_O3
             | LMN.LITELLM_OPENAI_O3_DEEP_RESEARCH
@@ -360,6 +373,10 @@ class LanguageModelInfo(BaseModel):
 
     supported_reasoning_efforts: list[ReasoningEffort] | None = None
 
+    # False for models that return HTTP 400 when tool_choice is a named tool or
+    # "any". Only "auto" and "none" are accepted there.
+    supports_forced_tool_choice: bool = True
+
     _ENV_VAR: ClassVar[str] = "LANGUAGE_MODEL_INFOS"
 
     def get_encoder(self) -> TypeEncoder:
@@ -1282,6 +1299,49 @@ class LanguageModelInfo(BaseModel):
                         "xhigh",
                     ],
                 )
+            case (
+                LanguageModelName.AZURE_GPT_6_LUNA_2026_0922
+                | LanguageModelName.AZURE_GPT_6_SOL_2026_0922
+            ):
+                info_cutoff_at = {
+                    LanguageModelName.AZURE_GPT_6_LUNA_2026_0922: date(2026, 5, 18),
+                    LanguageModelName.AZURE_GPT_6_SOL_2026_0922: date(2026, 4, 20),
+                }[model_name]
+                return cls(
+                    name=model_name,
+                    provider=LanguageModelProvider.AZURE,
+                    family=ModelFamily.OPENAI,
+                    version="2026-09-22",
+                    encoder_name=EncoderName.O200K_BASE,
+                    capabilities=[
+                        ModelCapabilities.CHAT_COMPLETIONS_API,
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.PARALLEL_FUNCTION_CALLING,
+                        ModelCapabilities.REASONING,
+                        ModelCapabilities.RESPONSES_API,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.STRUCTURED_OUTPUT,
+                        ModelCapabilities.VISION,
+                    ],
+                    token_limits=LanguageModelTokenLimits(
+                        token_limit_input=922_000, token_limit_output=128_000
+                    ),
+                    info_cutoff_at=info_cutoff_at,
+                    published_at=date(2026, 9, 22),
+                    temperature_bounds=TemperatureBounds(
+                        min_temperature=0.0, max_temperature=1.0
+                    ),
+                    default_options={
+                        "reasoning_effort": "medium",
+                    },
+                    supported_reasoning_efforts=[
+                        "none",
+                        "low",
+                        "medium",
+                        "high",
+                        "xhigh",
+                    ],
+                )
             case LanguageModelName.AZURE_GPT_4_TURBO_2024_0409:
                 return cls(
                     name=model_name,
@@ -1948,6 +2008,33 @@ class LanguageModelInfo(BaseModel):
                     published_at=date(2026, 7, 24),
                     supported_reasoning_efforts=[],
                 )
+            case (
+                LanguageModelName.ANTHROPIC_CLAUDE_OPUS_5_5
+                | LanguageModelName.VERTEX_CLAUDE_OPUS_5_5
+            ):
+                return cls(
+                    name=model_name,
+                    capabilities=[
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.VISION,
+                        ModelCapabilities.REASONING,
+                    ],
+                    provider=LanguageModelProvider.LITELLM,
+                    family=ModelFamily.ANTHROPIC,
+                    version="claude-opus-5-5",
+                    encoder_name=EncoderName.O200K_BASE,  # TODO: Update encoder with litellm
+                    token_limits=LanguageModelTokenLimits(
+                        # TODO: Remove the 1.3 adjustment once a proper Claude tokenizer is implemented. UN-24123
+                        # 1M context / 128K max output. https://platform.claude.com/docs/en/about-claude/models/overview
+                        token_limit_input=int(1_000_000 / 1.3),
+                        token_limit_output=int(128_000 / 1.3),
+                    ),
+                    info_cutoff_at=date(2026, 6, 1),
+                    published_at=date(2026, 9, 22),
+                    supported_reasoning_efforts=[],
+                    supports_forced_tool_choice=False,
+                )
             case (
                 LanguageModelName.ANTHROPIC_CLAUDE_FABLE_5
                 | LanguageModelName.VERTEX_CLAUDE_FABLE_5
@@ -2184,6 +2271,33 @@ class LanguageModelInfo(BaseModel):
                     published_at=date(2026, 7, 21),
                     supported_reasoning_efforts=["minimal", "low", "medium", "high"],
                 )
+            case LanguageModelName.GEMINI_3_8_FLASH:
+                return cls(
+                    name=model_name,
+                    capabilities=[
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.VISION,
+                        ModelCapabilities.STRUCTURED_OUTPUT,
+                        ModelCapabilities.REASONING,
+                    ],
+                    provider=LanguageModelProvider.LITELLM,
+                    family=ModelFamily.GOOGLE,
+                    version="gemini-3-8-flash",
+                    encoder_name=EncoderName.O200K_BASE,  # TODO: Update encoder with litellm
+                    token_limits=LanguageModelTokenLimits(
+                        # https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash
+                        token_limit_input=1_048_576,
+                        token_limit_output=65_536,
+                    ),
+                    # https://deepmind.google/models/model-cards/gemini-3-8-flash/
+                    info_cutoff_at=date(2026, 3, 1),
+                    # https://ai.google.dev/gemini-api/docs/changelog
+                    published_at=date(2026, 9, 2),
+                    # minimal is rejected; default thinking level is medium when unset.
+                    # https://ai.google.dev/gemini-api/docs/models/gemini-3.8-flash
+                    supported_reasoning_efforts=["low", "medium", "high"],
+                )
             case LanguageModelName.GEMINI_3_FLASH_PREVIEW:
                 return cls(
                     name=model_name,
@@ -2322,6 +2436,34 @@ class LanguageModelInfo(BaseModel):
                     supported_reasoning_efforts=["low", "medium", "high", "xhigh"],
                     default_options={"reasoning_effort": "high"},
                 )
+            case LanguageModelName.GROK_4_7:
+                # Specs from https://docs.x.ai/developers/models/grok-4.7
+                # Knowledge cutoff from https://docs.x.ai/developers/models
+                # Published 2026-09-21: https://vercel.com/changelog/grok-4-7-now-available-and-40-off-on-ai-gateway-fx-eve
+                return cls(
+                    name=model_name,
+                    capabilities=[
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.REASONING,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.STRUCTURED_OUTPUT,
+                        ModelCapabilities.VISION,
+                    ],
+                    provider=LanguageModelProvider.LITELLM,
+                    family=ModelFamily.XAI,
+                    version="grok-4.7",
+                    encoder_name=EncoderName.O200K_BASE,  # TODO: Update encoder with grok tokenizer
+                    token_limits=LanguageModelTokenLimits(
+                        # Context window is 500_000, we leave 20_000 tokens as buffer due to tokenizer mismatch
+                        # Assign 90% for input and 10% for output
+                        token_limit_input=int(480_000 * 0.9),
+                        token_limit_output=int(480_000 * 0.1),
+                    ),
+                    info_cutoff_at=date(2026, 5, day=1),
+                    published_at=date(2026, 9, day=21),
+                    supported_reasoning_efforts=["low", "medium", "high", "xhigh"],
+                    default_options={"reasoning_effort": "high"},
+                )
             case LanguageModelName.LITELLM_OPENAI_GPT_5:
                 return cls(
                     name=model_name,
@@ -2799,6 +2941,49 @@ class LanguageModelInfo(BaseModel):
                         "xhigh",
                     ],
                 )
+            case (
+                LanguageModelName.LITELLM_OPENAI_GPT_6_LUNA
+                | LanguageModelName.LITELLM_OPENAI_GPT_6_SOL
+            ):
+                info_cutoff_at = {
+                    LanguageModelName.LITELLM_OPENAI_GPT_6_LUNA: date(2026, 5, 18),
+                    LanguageModelName.LITELLM_OPENAI_GPT_6_SOL: date(2026, 4, 20),
+                }[model_name]
+                return cls(
+                    name=model_name,
+                    provider=LanguageModelProvider.LITELLM,
+                    family=ModelFamily.OPENAI,
+                    version="2026-09-22",
+                    encoder_name=EncoderName.O200K_BASE,
+                    capabilities=[
+                        ModelCapabilities.CHAT_COMPLETIONS_API,
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.PARALLEL_FUNCTION_CALLING,
+                        ModelCapabilities.REASONING,
+                        ModelCapabilities.RESPONSES_API,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.STRUCTURED_OUTPUT,
+                        ModelCapabilities.VISION,
+                    ],
+                    token_limits=LanguageModelTokenLimits(
+                        token_limit_input=922_000, token_limit_output=128_000
+                    ),
+                    info_cutoff_at=info_cutoff_at,
+                    published_at=date(2026, 9, 22),
+                    temperature_bounds=TemperatureBounds(
+                        min_temperature=0.0, max_temperature=1.0
+                    ),
+                    default_options={
+                        "reasoning_effort": "medium",
+                    },
+                    supported_reasoning_efforts=[
+                        "none",
+                        "low",
+                        "medium",
+                        "high",
+                        "xhigh",
+                    ],
+                )
             case LanguageModelName.LITELLM_OPENAI_O1:
                 return cls(
                     name=model_name,
@@ -3042,6 +3227,30 @@ class LanguageModelInfo(BaseModel):
                     published_at=date(2026, 1, 1),
                     supported_reasoning_efforts=[],
                 )
+            case LanguageModelName.LITELLM_DEEPSEEK_V4_1_FLASH:
+                return cls(
+                    name=model_name,
+                    provider=LanguageModelProvider.LITELLM,
+                    family=ModelFamily.DEEPSEEK,
+                    version="deepseek-v4-1-flash",
+                    encoder_name=EncoderName.DEEPSEEK,
+                    capabilities=[
+                        ModelCapabilities.FUNCTION_CALLING,
+                        ModelCapabilities.STRUCTURED_OUTPUT,
+                        ModelCapabilities.REASONING,
+                        ModelCapabilities.STREAMING,
+                        ModelCapabilities.VISION,
+                    ],
+                    # 1M context and 384K max output. Alibaba Model Studio lists
+                    # max input 1_000_000 and max output 393_216; Together serves
+                    # maxTokens 384_000.
+                    # https://www.alibabacloud.com/help/en/model-studio/deepseek-v4-1-flash
+                    # https://pi.dev/models/together/deepseek-ai-deepseek-v4-1-flash
+                    token_limits=LanguageModelTokenLimits(
+                        token_limit_input=1_000_000, token_limit_output=384_000
+                    ),
+                    supported_reasoning_efforts=[],
+                )
             case LanguageModelName.LITELLM_GLM_5_1:
                 return cls(
                     name=model_name,