New OpenAI GPT-5.1 (and 5.1 Codex) models

[simonw]

• https://platform.openai.com/docs/models/gpt-5.1
• https://platform.openai.com/docs/models/gpt-5.1-chat-latest
• https://platform.openai.com/docs/models/gpt-5.1-codex
• https://platform.openai.com/docs/models/gpt-5.1-codex-mini

[simonw]

diff --git a/llm/default_plugins/openai_models.py b/llm/default_plugins/openai_models.py
index 94c1ffc..2c10877 100644
--- a/llm/default_plugins/openai_models.py
+++ b/llm/default_plugins/openai_models.py
@@ -182,6 +182,30 @@ def register_models(register):
                 supports_tools=True,
             ),
         )
+    # GPT-5.1
+    for model_id in (
+        "gpt-5.1",
+        "gpt-5.1-chat-latest",
+        "gpt-5.1-codex",
+        "gpt-5.1-codex-mini",
+    ):
+        register(
+            Chat(
+                model_id,
+                vision=True,
+                reasoning=True,
+                supports_schema=True,
+                supports_tools=True,
+            ),
+            AsyncChat(
+                model_id,
+                vision=True,
+                reasoning=True,
+                supports_schema=True,
+                supports_tools=True,
+            ),
+        )
+
     # The -instruct completion model
     register(
         Completion("gpt-3.5-turbo-instruct", default_max_tokens=256),

[simonw]

uv run llm -m gpt-5.1 -t pelican-svg
uv run llm -m gpt-5.1-chat-latest -t pelican-svg
uv run llm -m gpt-5.1-codex -t pelican-svg
uv run llm -m gpt-5.1-codex-mini -t pelican-svg

The first two worked: https://gist.github.com/simonw/fce773879bcd66e01df59a5011899532

It looks like the codex models require the Responses API.

[simonw]

Here's the GPT-5.1 pelican:

And for gpt-5.1-chat-latest:

[simonw]

Tried this instead:

uv run llm -m gpt-5.1 -t pelican-svg -o reasoning_effort high

https://gist.github.com/simonw/fd120c35d8d34512c6f15d8d8e5192ec

[simonw]

Just spotted this in https://openai.com/index/gpt-5-1-for-developers/

Developers can now use GPT‑5.1 without reasoning by setting reasoning_effort to 'none'.

Need to add that to this enum:

https://github.com/simonw/llm/blob/0526abeeea021b23552bcdafc15c278f2bb2e119/llm/default_plugins/openai_models.py#L456-L460

... or maybe we don't need to change that, since the default for 5.1 is none if you don't specify one.

[simonw]

Need to add support for extended prompt cache retention: https://platform.openai.com/docs/guides/prompt-caching#extended-prompt-cache-retention

To use extended caching with GPT‑5.1, add the parameter "prompt_cache_retention='24h'" on the Responses or Chat Completions API. See the prompt caching docs⁠ for more detail.

[liammonahan]

What's a good next step here? The 5.1 chat (non-codex) models seem to be in good shape. Maybe pushing a release with those could unblock users while continuing to work on the codex integration? Thanks for all your work on this library.