From 21d08c8482104d672eaf92ce4e3721b6e2c627d8 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sun, 1 Nov 2026 09:14:06 -0800
Subject: [PATCH] fix(anthropic): decode text attachments under every WHATWG
 label Python knows

The allowlist kept Python codec names for one label per WHATWG encoding, so
labels that Python maps to a codec of its own (iso-8859-9, iso-8859-11,
tis-620, gb2312, big5-hkscs, ms932) answered 400 instead of decoding. All six
are linear table codecs, so the punycode and utf-7 rejection stands
---
 .../litellm_core_utils/prompt_templates/factory.py  |  6 +++++-
 ...t_litellm_core_utils_prompt_templates_factory.py | 13 ++++++++++++-
 2 files changed, 17 insertions(+), 2 deletions(-)

diff --git a/litellm/litellm_core_utils/prompt_templates/factory.py b/litellm/litellm_core_utils/prompt_templates/factory.py
index 1170dca53d8..3ee61a9a94a 100644
--- a/litellm/litellm_core_utils/prompt_templates/factory.py
+++ b/litellm/litellm_core_utils/prompt_templates/factory.py
@@ -1868,18 +1868,22 @@ _TEXT_ATTACHMENT_ENCODINGS: Final = frozenset(
         "utf-16-be",
         "ascii",
         "latin-1",
-        *(f"iso8859-{part}" for part in (2, 3, 4, 5, 6, 7, 8, 10, 13, 14, 15, 16)),
+        *(f"iso8859-{part}" for part in (*range(2, 12), *range(13, 17))),
+        "tis-620",
         *(f"cp{code_page}" for code_page in (866, 874, *range(1250, 1259))),
         "koi8-r",
         "koi8-u",
         "mac-roman",
         "mac-cyrillic",
+        "gb2312",
         "gbk",
         "gb18030",
         "big5",
+        "big5hkscs",
         "euc-jp",
         "iso2022-jp",
         "shift-jis",
+        "cp932",
         "euc-kr",
     )
 )
diff --git a/tests/unit/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py b/tests/unit/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
index a3e802e97a4..c8d53c12aaf 100644
--- a/tests/unit/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
+++ b/tests/unit/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py
@@ -9383,7 +9383,18 @@ def test_anthropic_messages_pt_text_document_keeps_charset_named_by_the_data_uri
 
 @pytest.mark.parametrize(
     ("charset", "text"),
-    [("shift_jis", "日本語"), ("windows-1252", "café"), ("KOI8-R", "привет"), ("utf-16", "naïve")],
+    [
+        ("shift_jis", "日本語"),
+        ("windows-1252", "café"),
+        ("KOI8-R", "привет"),
+        ("utf-16", "naïve"),
+        ("iso-8859-9", "İstanbul"),
+        ("iso-8859-11", "ไทย"),
+        ("tis-620", "ไทย"),
+        ("gb2312", "中文"),
+        ("big5-hkscs", "香港"),
+        ("ms932", "日本語"),
+    ],
 )
 def test_anthropic_messages_pt_reads_a_text_attachment_in_a_legacy_encoding_its_data_uri_names(charset, text):
     data_uri: Final = f"data:text/plain;charset={charset};base64," + base64.b64encode(text.encode(charset)).decode()