release: 2.12.0

.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "2.11.0"
+  ".": "2.12.0"
 }

.stats.yml

@@ -1,4 +1,4 @@
 configured_endpoints: 75
-openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai%2Ftogetherai-19148ebeed55db4be0c1c60d31c81306386c4c71a2aee8d1eeac8ebfa15c1168.yml
-openapi_spec_hash: 747224d60adf1ab51760e5752ce1b846
-config_hash: 279a64a8df3e6dba241bfa1e18a04f69
+openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/togetherai/togetherai-e1200616b1a93d40e478800d2c9e06ddeb10b508f2a9aa65810ee31878ba4f23.yml
+openapi_spec_hash: 23245993d115722da1b697f10799f4f1
+config_hash: 6c214c91fad5ead4849be777fd9e8108

CHANGELOG.md

@@ -1,5 +1,39 @@
 # Changelog
 
+## 2.12.0 (2026-05-01)
+
+Full Changelog: [v2.11.0...v2.12.0](https://github.com/togethercomputer/together-py/compare/v2.11.0...v2.12.0)
+
+### Features
+
+* Add default retrieve handler when passing an id to commands ([#345](https://github.com/togethercomputer/together-py/issues/345)) ([984519c](https://github.com/togethercomputer/together-py/commit/984519c038d4b99a794b5665c5c74b7c22f8802c))
+* add the pronunciation dict ([f59d672](https://github.com/togethercomputer/together-py/commit/f59d672d1f26abc9b4ad6216209b6d6707cfa88b))
+* **api:** add cached_input field to Pricing model ([c03fb18](https://github.com/togethercomputer/together-py/commit/c03fb180dc52531e4c99ebcd2a2ca8acea81dc9b))
+* **api:** Update server url to .ai ([2ea6a2d](https://github.com/togethercomputer/together-py/commit/2ea6a2d9201373c8959b4c34749236183a785543))
+* ENG-87042: clarify TTS language parameter supports lowercase locales ([95cf65b](https://github.com/togethercomputer/together-py/commit/95cf65b30233791b1e73f823a3cb0e6cdb731ab4))
+* ENG-87042: document language on TTS WebSocket and simplify locale note ([14b429f](https://github.com/togethercomputer/together-py/commit/14b429f0e08c8ad9eba4ffb51ba6d8d95d35973d))
+* MOSH-2181: Add default note on max-seq-length ([b1edd58](https://github.com/togethercomputer/together-py/commit/b1edd58f1bc7ccc2f3b1016ac5d09eaf2fad16b3))
+* Show cache input token pricing in models list cli output ([#350](https://github.com/togethercomputer/together-py/issues/350)) ([1476f70](https://github.com/togethercomputer/together-py/commit/1476f709115b23e6ca08468c4b0ca6e72553d967))
+
+
+### Bug Fixes
+
+* Properly handle unlinking temp file during fine-tuning download ([#352](https://github.com/togethercomputer/together-py/issues/352)) ([2f1c458](https://github.com/togethercomputer/together-py/commit/2f1c458ee388a14757c50b077e8402c0a063fe01))
+* **types:** remove eval-sample, eval-output, eval-summary, batch-generated from FilePurpose ([ea1d61a](https://github.com/togethercomputer/together-py/commit/ea1d61a182cad4f11f0e123a23381284f61b2153))
+
+
+### Chores
+
+* Improve --json support accross multiple commands ([#347](https://github.com/togethercomputer/together-py/issues/347)) ([949bfa4](https://github.com/togethercomputer/together-py/commit/949bfa4514e34b761727622fab3d345e940a7fe9))
+* **internal:** reformat pyproject.toml ([a52e64b](https://github.com/togethercomputer/together-py/commit/a52e64bde79ccfe3f06c9d4a59af5e736889e412))
+* Pass agent name in request headers when detected ([#351](https://github.com/togethercomputer/together-py/issues/351)) ([1668cf4](https://github.com/togethercomputer/together-py/commit/1668cf4208e4c32c7ee91a6978b708bc8c4c8256))
+* Update references from .xyz to .ai domain ([#349](https://github.com/togethercomputer/together-py/issues/349)) ([0c33450](https://github.com/togethercomputer/together-py/commit/0c334506cfb7f56a7cc8fb0bba4c4b89aed0b408))
+
+
+### Documentation
+
+* polish CLI help text; introduce Options + Global Options panels ([#346](https://github.com/togethercomputer/together-py/issues/346)) ([4905c1a](https://github.com/togethercomputer/together-py/commit/4905c1ac807de4b317639524720592d51b1d23f9))
+
 ## 2.11.0 (2026-04-28)
 
 Full Changelog: [v2.10.0...v2.11.0](https://github.com/togethercomputer/together-py/compare/v2.10.0...v2.11.0)

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "together"
-version = "2.11.0"
+version = "2.12.0"
 description = "The official Python library for the together API"
 dynamic = ["readme"]
 license = "Apache-2.0"
@@ -177,7 +177,7 @@ show_error_codes = true
 #
 # We also exclude our `tests` as mypy doesn't always infer
 # types correctly and Pyright will still catch any type errors.
-exclude = ['src/together/_files.py', '_dev/.*.py', 'tests/.*']
+exclude = ["src/together/_files.py", "_dev/.*.py", "tests/.*"]
 
 strict_equality = true
 implicit_reexport = true

src/together/_base_client.py

@@ -38,6 +38,7 @@
 import pydantic
 from httpx import URL
 from pydantic import PrivateAttr
+from detect_agent import determine_agent
 
 from . import _exceptions
 from ._qs import Querystring
@@ -675,6 +676,7 @@ def default_headers(self) -> dict[str, str | Omit]:
             "Content-Type": "application/json",
             "User-Agent": self.user_agent,
             **self.platform_headers(),
+            **self.agent_headers(),
             **self.auth_headers,
             **self._custom_headers,
         }
@@ -714,6 +716,12 @@ def platform_headers(self) -> Dict[str, str]:
         # https://github.com/python/cpython/issues/88476
         return platform_headers(self._version, platform=self._platform)
 
+    def agent_headers(self) -> dict[str, str]:
+        # the actual implementation is in a separate `lru_cache` decorated
+        # function because adding `lru_cache` to methods will leak memory
+        # https://github.com/python/cpython/issues/88476
+        return agent_headers()
+
     def _parse_retry_after_header(self, response_headers: Optional[httpx.Headers] = None) -> float | None:
         """Returns a float of the number of seconds (not milliseconds) to wait after retrying, or None if unspecified.
 
@@ -2055,6 +2063,15 @@ def get_platform() -> Platform:
     return "Unknown"
 
 
+@lru_cache(maxsize=None)
+def agent_headers() -> dict[str, str]:
+    agent = determine_agent()
+    if agent["is_agent"] and agent["agent"]:
+        return {"X-Stainless-Agent": agent["agent"]["name"]}
+
+    return {}
+
+
 @lru_cache(maxsize=None)
 def platform_headers(version: str, *, platform: Platform | None) -> Dict[str, str]:
     return {

src/together/_client.py

@@ -128,7 +128,7 @@ def __init__(
             base_url = os.environ.get("TOGETHER_BASE_URL")
         self._base_url_overridden = base_url is not None
         if base_url is None:
-            base_url = f"https://api.together.xyz/v1"
+            base_url = f"https://api.together.ai/v1"
 
         custom_headers_env = os.environ.get("TOGETHER_CUSTOM_HEADERS")
         if custom_headers_env is not None:
@@ -402,7 +402,7 @@ def __init__(
             base_url = os.environ.get("TOGETHER_BASE_URL")
         self._base_url_overridden = base_url is not None
         if base_url is None:
-            base_url = f"https://api.together.xyz/v1"
+            base_url = f"https://api.together.ai/v1"
 
         custom_headers_env = os.environ.get("TOGETHER_CUSTOM_HEADERS")
         if custom_headers_env is not None:

src/together/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "together"
-__version__ = "2.11.0"  # x-release-please-version
+__version__ = "2.12.0"  # x-release-please-version

src/together/constants.py

@@ -31,4 +31,4 @@
 MAX_CONNECTION_RETRIES = 2
 MAX_RETRIES = DEFAULT_MAX_RETRIES
 
-BASE_URL = "https://api.together.xyz/v1"
+BASE_URL = "https://api.together.ai/v1"