branch: elpa/gptel
commit 34ecadea17870b07fe7f146376b5a25f45fb27af
Author: Karthik Chikmagalur <karthikchikmaga...@gmail.com>
Commit: Karthik Chikmagalur <karthikchikmaga...@gmail.com>
gptel-anthropic: Update cache control to include ttl "1h"
* README.org ((Optional) Interim support for Claude 3.7 Sonnet):
Simplify the setup for Claude-thinking. None of the beta headers
are required any longer.
* gptel-anthropic.el (gptel--request-data, gptel--parse-list,
gptel--parse-buffer): Increase default cache ttl for system and
messages to 1 hour. Tools cache remains at a ttl of 5 minutes.
It's not clear from the documentation if 1 hour caches are any
more expensive than 5 minute caches, but on the whole this should
reduce the costs of conversations when using `gptel-cache' with
Anthropic (only).
(gptel-make-anthropic): Remove obsolete headers, add
extended-cache-ttl support.
---
README.org | 8 +-------
gptel-anthropic.el | 12 ++++++------
2 files changed, 7 insertions(+), 13 deletions(-)
diff --git a/README.org b/README.org
index 349cdaae27..bec5645414 100644
--- a/README.org
+++ b/README.org
@@ -615,13 +615,7 @@ To use Claude 3.7 Sonnet model in its "thinking" mode, you
can define a second C
(gptel-make-anthropic "Claude-thinking" ;Any name you want
:key "your-API-key"
:stream t
- :models '(claude-3-7-sonnet-20250219)
- :header (lambda () (when-let* ((key (gptel--get-api-key)))
- `(("x-api-key" . ,key)
- ("anthropic-version" . "2023-06-01")
- ("anthropic-beta" . "pdfs-2024-09-25")
- ("anthropic-beta" . "output-128k-2025-02-19")
- ("anthropic-beta" . "prompt-caching-2024-07-31"))))
+ :models '(claude-sonnet-4-20250514 claude-3-7-sonnet-20250219)
:request-params '(:thinking (:type "enabled" :budget_tokens 2048)
:max_tokens 4096))
#+end_src
diff --git a/gptel-anthropic.el b/gptel-anthropic.el
index 181e79f309..ed47d32370 100644
--- a/gptel-anthropic.el
+++ b/gptel-anthropic.el
@@ -216,7 +216,7 @@ Mutate state INFO with response metadata."
;; gptel--system-message is guaranteed to be a string
(plist-put prompts-plist :system
`[(:type "text" :text ,gptel--system-message
- :cache_control (:type "ephemeral"))])
+ :cache_control (:type "ephemeral" :ttl "1h"))])
(plist-put prompts-plist :system gptel--system-message)))
(when gptel-temperature
(plist-put prompts-plist :temperature gptel-temperature))
@@ -362,7 +362,7 @@ TOOL-USE is a list of plists containing tool names,
arguments and call results."
(when (and (or (eq gptel-cache t) (memq 'message gptel-cache))
(gptel--model-capable-p 'cache))
(nconc (aref (plist-get (car (last full-prompt)) :content) 0)
- '(:cache_control (:type "ephemeral"))))
+ '(:cache_control (:type "ephemeral" :ttl "1h"))))
full-prompt))
(cl-defmethod gptel--parse-buffer ((backend gptel-anthropic) &optional
max-entries)
@@ -429,9 +429,10 @@ TOOL-USE is a list of plists containing tool names,
arguments and call results."
(if (stringp last-message)
(plist-put
(car (last prompts)) :content
- `[(:type "text" :text ,last-message :cache_control (:type
"ephemeral"))])
+ `[(:type "text" :text ,last-message
+ :cache_control (:type "ephemeral" :ttl "1h"))])
(nconc (aref (plist-get (car (last prompts)) :content) 0)
- '(:cache_control (:type "ephemeral"))))))
+ '(:cache_control (:type "ephemeral" :ttl "1h"))))))
prompts))
(defun gptel--anthropic-parse-multipart (parts)
@@ -621,8 +622,7 @@ URL
`https://docs.anthropic.com/en/docs/about-claude/models#model-comparison-tab
(lambda () (when-let* ((key (gptel--get-api-key)))
`(("x-api-key" . ,key)
("anthropic-version" . "2023-06-01")
- ("anthropic-beta" . "pdfs-2024-09-25")
- ("anthropic-beta" . "prompt-caching-2024-07-31")))))
+ ("anthropic-beta" . "extended-cache-ttl-2025-04-11")))))
(models gptel--anthropic-models)
(host "api.anthropic.com")
(protocol "https")
