From bca80d41d98350a8be3c254c92721b53a68b1833 Mon Sep 17 00:00:00 2001 From: Yishuai Li Date: Tue, 15 Sep 2026 16:16:24 +0800 Subject: [PATCH 1/3] Add kimi-k3 to alibaba and alibaba-cn providers --- providers/alibaba-cn/models/kimi-k3.toml | 22 ++++++++++++++++++++++ providers/alibaba/models/kimi-k3.toml | 22 ++++++++++++++++++++++ 2 files changed, 44 insertions(+) create mode 100644 providers/alibaba-cn/models/kimi-k3.toml create mode 100644 providers/alibaba/models/kimi-k3.toml diff --git a/providers/alibaba-cn/models/kimi-k3.toml b/providers/alibaba-cn/models/kimi-k3.toml new file mode 100644 index 00000000000..23ace1c5962 --- /dev/null +++ b/providers/alibaba-cn/models/kimi-k3.toml @@ -0,0 +1,22 @@ +# Sources (accessed 2026-09-15): +# https://www.alibabacloud.com/help/en/model-studio/kimi-k3 (China/Beijing USD +# pricing, limits, capability table) +# https://www.alibabacloud.com/help/en/model-studio/kimi-api (thinking-only: +# enable_thinking cannot be disabled; no thinking_budget; no reasoning_effort; +# text+image input only, no video; reasoning_content streamed in deltas) +base_model = "moonshotai/kimi-k3" +reasoning_options = [] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 2.827 +output = 14.133 +cache_read = 0.283 + +[modalities] +input = ["text", "image"] + +[limit] +output = 1_048_576 diff --git a/providers/alibaba/models/kimi-k3.toml b/providers/alibaba/models/kimi-k3.toml new file mode 100644 index 00000000000..0c327d6ea0f --- /dev/null +++ b/providers/alibaba/models/kimi-k3.toml @@ -0,0 +1,22 @@ +# Sources (accessed 2026-09-15): +# https://www.qwencloud.com/models/kimi-k3 (Singapore/international pricing, limits) +# https://www.alibabacloud.com/help/en/model-studio/kimi-k3 (limits, capability table) +# https://www.alibabacloud.com/help/en/model-studio/kimi-api (thinking-only: +# enable_thinking cannot be disabled; no thinking_budget; no reasoning_effort; +# text+image input only, no video; reasoning_content streamed in deltas) +base_model = "moonshotai/kimi-k3" +reasoning_options = [] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 3 +output = 15 +cache_read = 0.3 + +[modalities] +input = ["text", "image"] + +[limit] +output = 1_048_576 From fa8cbe09e67f680f2a59954a5e98e87a4bac1bbe Mon Sep 17 00:00:00 2001 From: Yishuai Li Date: Tue, 15 Sep 2026 16:21:08 +0800 Subject: [PATCH 2/3] Fix kimi-k3 reasoning options: effort low|high|max per zh API reference --- providers/alibaba-cn/models/kimi-k3.toml | 10 ++++++---- providers/alibaba/models/kimi-k3.toml | 10 ++++++---- 2 files changed, 12 insertions(+), 8 deletions(-) diff --git a/providers/alibaba-cn/models/kimi-k3.toml b/providers/alibaba-cn/models/kimi-k3.toml index 23ace1c5962..966fa8014ae 100644 --- a/providers/alibaba-cn/models/kimi-k3.toml +++ b/providers/alibaba-cn/models/kimi-k3.toml @@ -1,11 +1,13 @@ # Sources (accessed 2026-09-15): # https://www.alibabacloud.com/help/en/model-studio/kimi-k3 (China/Beijing USD # pricing, limits, capability table) -# https://www.alibabacloud.com/help/en/model-studio/kimi-api (thinking-only: -# enable_thinking cannot be disabled; no thinking_budget; no reasoning_effort; -# text+image input only, no video; reasoning_content streamed in deltas) +# https://www.alibabacloud.com/help/zh/model-studio/qwen-api-via-openai-chat-completions +# (reasoning_effort) and /help/en/model-studio/kimi-api (text+image input only, +# no video; reasoning_content streamed in deltas) +# Effort: reasoning_effort = low|high|max (default max). Thinking is always on: +# enable_thinking only accepts true (passing false errors), no thinking_budget. base_model = "moonshotai/kimi-k3" -reasoning_options = [] +reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] [interleaved] field = "reasoning_content" diff --git a/providers/alibaba/models/kimi-k3.toml b/providers/alibaba/models/kimi-k3.toml index 0c327d6ea0f..07bcf879841 100644 --- a/providers/alibaba/models/kimi-k3.toml +++ b/providers/alibaba/models/kimi-k3.toml @@ -1,11 +1,13 @@ # Sources (accessed 2026-09-15): # https://www.qwencloud.com/models/kimi-k3 (Singapore/international pricing, limits) # https://www.alibabacloud.com/help/en/model-studio/kimi-k3 (limits, capability table) -# https://www.alibabacloud.com/help/en/model-studio/kimi-api (thinking-only: -# enable_thinking cannot be disabled; no thinking_budget; no reasoning_effort; -# text+image input only, no video; reasoning_content streamed in deltas) +# https://www.alibabacloud.com/help/zh/model-studio/qwen-api-via-openai-chat-completions +# (reasoning_effort) and /help/en/model-studio/kimi-api (text+image input only, +# no video; reasoning_content streamed in deltas) +# Effort: reasoning_effort = low|high|max (default max). Thinking is always on: +# enable_thinking only accepts true (passing false errors), no thinking_budget. base_model = "moonshotai/kimi-k3" -reasoning_options = [] +reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] [interleaved] field = "reasoning_content" From a848ac0f38a43937461a5739acd7d8ba01165d79 Mon Sep 17 00:00:00 2001 From: Yishuai Li Date: Tue, 15 Sep 2026 16:31:36 +0800 Subject: [PATCH 3/3] Use block-style reasoning_options, matching moonshotai/kimi-k3 style --- providers/alibaba-cn/models/kimi-k3.toml | 5 ++++- providers/alibaba/models/kimi-k3.toml | 9 ++++++--- 2 files changed, 10 insertions(+), 4 deletions(-) diff --git a/providers/alibaba-cn/models/kimi-k3.toml b/providers/alibaba-cn/models/kimi-k3.toml index 966fa8014ae..0df890f8e11 100644 --- a/providers/alibaba-cn/models/kimi-k3.toml +++ b/providers/alibaba-cn/models/kimi-k3.toml @@ -7,7 +7,10 @@ # Effort: reasoning_effort = low|high|max (default max). Thinking is always on: # enable_thinking only accepts true (passing false errors), no thinking_budget. base_model = "moonshotai/kimi-k3" -reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] [interleaved] field = "reasoning_content" diff --git a/providers/alibaba/models/kimi-k3.toml b/providers/alibaba/models/kimi-k3.toml index 07bcf879841..28505229989 100644 --- a/providers/alibaba/models/kimi-k3.toml +++ b/providers/alibaba/models/kimi-k3.toml @@ -7,14 +7,17 @@ # Effort: reasoning_effort = low|high|max (default max). Thinking is always on: # enable_thinking only accepts true (passing false errors), no thinking_budget. base_model = "moonshotai/kimi-k3" -reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] [interleaved] field = "reasoning_content" [cost] -input = 3 -output = 15 +input = 3.0 +output = 15.0 cache_read = 0.3 [modalities]