From f6257033f06f52e4b898cbab7476f0cf32b985f2 Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 12:21:40 -0700 Subject: [PATCH 1/7] fix(cloudflare): refresh catalogue reasoning and canonical links --- .../aisingapore/gemma-sea-lion-v4-27b-it.toml | 17 +------------ .../deepseek-ai/deepseek-v4-flash-0731.toml | 9 +++---- .../@cf/deepseek-ai/deepseek-v4-pro-0813.toml | 8 +++--- .../models/@cf/google/gemma-4-26b-a4b-it.toml | 7 +++--- .../@cf/ibm-granite/granite-4.0-h-micro.toml | 13 +--------- .../mistral-small-3.1-24b-instruct.toml | 10 +------- .../models/@cf/moonshotai/kimi-k2.6.toml | 7 +++--- .../models/@cf/moonshotai/kimi-k2.7-code.toml | 4 ++- .../models/@cf/zai-org/glm-4.7-flash.toml | 25 ++++++------------- .../models/@cf/zai-org/glm-5.2.toml | 7 +++--- .../models/@cf/zai-org/glm-5.3-flash.toml | 18 +++---------- .../models/@cf/zai-org/glm-5.3.toml | 4 ++- 12 files changed, 38 insertions(+), 91 deletions(-) diff --git a/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml b/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml index 8976b92b55a..8e5628d902d 100644 --- a/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml +++ b/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml @@ -1,23 +1,8 @@ +base_model = "aisingapore/gemma-sea-lion-v4-27b-it" name = "Gemma Sea Lion V4 27B It" description = "Open Gemma instruction model for efficient chat and self-hosted deployments" -family = "gemma" -release_date = "2025-09-23" -last_updated = "2025-09-23" -attachment = false -reasoning = false -temperature = true -tool_call = false structured_output = false -open_weights = true [cost] input = 0.351 output = 0.555 - -[limit] -context = 128_000 -output = 128_000 - -[modalities] -input = ["text"] -output = ["text"] diff --git a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml index b72f1489e71..84b4fc57fd4 100644 --- a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml +++ b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml @@ -1,13 +1,10 @@ -# Toggle: thinking.type = enabled|disabled. -# Effort: reasoning_effort = high|max. +# Effort: reasoning_effort = none|low|high|max; none disables thinking. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "deepseek/deepseek-v4-flash-0731" -[[reasoning_options]] -type = "toggle" - [[reasoning_options]] type = "effort" -values = ["high", "max"] +values = ["none", "low", "high", "max"] [cost] input = 0.44 diff --git a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml index b28c06ffbe1..262c61fe520 100644 --- a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml +++ b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml @@ -1,8 +1,10 @@ -# Toggle: thinking.type = enabled|disabled. -# Effort: reasoning_effort = high|max. +# Effort: reasoning_effort = none|low|high|max; none disables thinking. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "deepseek/deepseek-v4-pro-0813" -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["high", "max"] }] +[[reasoning_options]] +type = "effort" +values = ["none", "low", "high", "max"] [cost] input = 1.32 diff --git a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml index 817334b2cee..bf52e85efce 100644 --- a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml +++ b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml @@ -1,9 +1,8 @@ -# Native `/ai/run` accepts `reasoning_effort = low|medium|high` and -# `chat_template_kwargs.enable_thinking = true|false`; no budget is documented. -# https://developers.cloudflare.com/workers-ai/models/gemma-4-26b-a4b-it/sync-input.json (accessed 2026-06-25) +# Effort: reasoning_effort = none|high; none disables thinking. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "google/gemma-4-26b-a4b-it" -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }] +reasoning_options = [{ type = "effort", values = ["none", "high"] }] interleaved = true [cost] diff --git a/providers/cloudflare-workers-ai/models/@cf/ibm-granite/granite-4.0-h-micro.toml b/providers/cloudflare-workers-ai/models/@cf/ibm-granite/granite-4.0-h-micro.toml index a43a661e427..1aa972501cf 100644 --- a/providers/cloudflare-workers-ai/models/@cf/ibm-granite/granite-4.0-h-micro.toml +++ b/providers/cloudflare-workers-ai/models/@cf/ibm-granite/granite-4.0-h-micro.toml @@ -1,14 +1,7 @@ +base_model = "ibm/granite-4-h-micro" name = "Granite 4.0 H Micro" description = "Efficient model for low-latency assistance, extraction, and routine automation" -family = "granite" -release_date = "2025-10-07" -last_updated = "2025-10-07" -attachment = false -reasoning = false -temperature = true -tool_call = true structured_output = false -open_weights = true [cost] input = 0.017 @@ -17,7 +10,3 @@ output = 0.112 [limit] context = 131_000 output = 131_000 - -[modalities] -input = ["text"] -output = ["text"] diff --git a/providers/cloudflare-workers-ai/models/@cf/mistralai/mistral-small-3.1-24b-instruct.toml b/providers/cloudflare-workers-ai/models/@cf/mistralai/mistral-small-3.1-24b-instruct.toml index e456a0e8625..7ed149c081e 100644 --- a/providers/cloudflare-workers-ai/models/@cf/mistralai/mistral-small-3.1-24b-instruct.toml +++ b/providers/cloudflare-workers-ai/models/@cf/mistralai/mistral-small-3.1-24b-instruct.toml @@ -1,23 +1,15 @@ +base_model = "mistral/mistral-small-3-1-24b-instruct-2503" name = "Mistral Small 3.1 24B Instruct" description = "Efficient Mistral model for fast chat, extraction, and production assistants" -family = "mistral-small" -release_date = "2025-03-18" -last_updated = "2025-03-18" attachment = false -reasoning = false -temperature = true -tool_call = true structured_output = false -open_weights = true [cost] input = 0.351 output = 0.555 [limit] -context = 128_000 output = 128_000 [modalities] input = ["text"] -output = ["text"] diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml index 3c4c22a8305..fb8d1193ce2 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml @@ -1,9 +1,8 @@ -# Native `/ai/run` accepts `reasoning_effort = low|medium|high` and -# `chat_template_kwargs.thinking = true|false`; no budget is documented. -# https://developers.cloudflare.com/workers-ai/models/kimi-k2.6/sync-input.json (accessed 2026-06-25) +# Effort: reasoning_effort = none|high; none selects instant mode. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "moonshotai/kimi-k2.6" -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }] +reasoning_options = [{ type = "effort", values = ["none", "high"] }] [interleaved] field = "reasoning_content" diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml index a06c8bc9a29..23333264c83 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml @@ -1,5 +1,7 @@ +# Effort: reasoning_effort = high; thinking is mandatory. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "moonshotai/kimi-k2.7-code" -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }] +reasoning_options = [{ type = "effort", values = ["high"] }] temperature = true [cost] diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml index 197fa7c94b5..e2b003ca0af 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml @@ -1,32 +1,23 @@ # Native `/ai/run` accepts `reasoning_effort = low|medium|high` and # `chat_template_kwargs.enable_thinking = true|false`; no budget is documented. # https://developers.cloudflare.com/workers-ai/models/glm-4.7-flash/sync-input.json (accessed 2026-06-25) - -name = "GLM-4.7-Flash" +base_model = "zhipuai/glm-4.7-flash" description = "Efficient GLM model for fast reasoning, coding, and agent workflows" -family = "glm-flash" -release_date = "2026-01-19" -last_updated = "2026-01-19" -attachment = false -reasoning = true -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }] -temperature = true -tool_call = true structured_output = true -knowledge = "2025-04" -open_weights = true [interleaved] field = "reasoning_content" +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "medium", "high"] + [cost] input = 0.0605 output = 0.4 [limit] context = 131_072 -output = 131_072 - -[modalities] -input = ["text"] -output = ["text"] diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.2.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.2.toml index 5642ab6e4ea..48337ad55f9 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.2.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.2.toml @@ -1,14 +1,13 @@ +# Effort: reasoning_effort = none|high|max; none disables thinking. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). # Context and output limits verified against Cloudflare Workers AI's OpenAI-compatible endpoint. # https://developers.cloudflare.com/workers-ai/models/glm-5.2/ base_model = "zhipuai/glm-5.2" name = "Glm 5.2" -[[reasoning_options]] -type = "toggle" - [[reasoning_options]] type = "effort" -values = ["low", "medium", "high"] +values = ["none", "high", "max"] [cost] input = 1.4 diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml index 000ebf8a49a..968bc3c9482 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml @@ -1,18 +1,9 @@ -# Context/output capped to Cloudflare Workers AI context window (1,048,576). -# Catalog/search previously reported 1,310,720, which exceeds AI Gateway max_completion_tokens. -# https://developers.cloudflare.com/workers-ai/models/glm-5.3-flash/ +# Effort: reasoning_effort = low|high|max; thinking is mandatory. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +base_model = "zhipuai/glm-5.3-flash" name = "Glm 5.3 Flash" description = "GLM vision model for visual reasoning, documents, and multimodal agents" -family = "glm-flash" -release_date = "2026-08-26" -last_updated = "2026-08-26" -attachment = true -reasoning = true -temperature = true -tool_call = true -structured_output = true -open_weights = true -reasoning_options = [] +reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] [cost] input = 0.15 @@ -25,4 +16,3 @@ output = 1_048_576 [modalities] input = ["text", "image"] -output = ["text"] diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml index a5f20bbaf80..52edd473a26 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml @@ -1,7 +1,9 @@ +# Effort: reasoning_effort = low|high|max; thinking is mandatory. +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "zhipuai/glm-5.3" name = "Glm 5.3" description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" -reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }] +reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] [cost] input = 1.4 From 8c47902e2927cd1d5649ab7c3c0c9863ae2e9d5e Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 13:39:17 -0700 Subject: [PATCH 2/7] fix(cloudflare): sync GLM 4.7 toggle and expand reasoning tables --- .../models/@cf/google/gemma-4-26b-a4b-it.toml | 5 ++++- .../models/@cf/moonshotai/kimi-k2.6.toml | 5 ++++- .../models/@cf/moonshotai/kimi-k2.7-code.toml | 5 ++++- .../models/@cf/nvidia/nemotron-3-120b-a12b.toml | 8 +++++++- .../models/@cf/qwen/qwen3.8-27b.toml | 8 +++++++- .../models/@cf/zai-org/glm-4.7-flash.toml | 9 ++------- .../models/@cf/zai-org/glm-5.3-flash.toml | 5 ++++- .../models/@cf/zai-org/glm-5.3.toml | 5 ++++- 8 files changed, 36 insertions(+), 14 deletions(-) diff --git a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml index bf52e85efce..d2790a7f64a 100644 --- a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml +++ b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml @@ -2,9 +2,12 @@ # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "google/gemma-4-26b-a4b-it" -reasoning_options = [{ type = "effort", values = ["none", "high"] }] interleaved = true +[[reasoning_options]] +type = "effort" +values = ["none", "high"] + [cost] input = 0.1 output = 0.3 diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml index fb8d1193ce2..d3f174f1a8e 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml @@ -2,7 +2,10 @@ # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "moonshotai/kimi-k2.6" -reasoning_options = [{ type = "effort", values = ["none", "high"] }] + +[[reasoning_options]] +type = "effort" +values = ["none", "high"] [interleaved] field = "reasoning_content" diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml index 23333264c83..79685ac02e4 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml @@ -1,9 +1,12 @@ # Effort: reasoning_effort = high; thinking is mandatory. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "moonshotai/kimi-k2.7-code" -reasoning_options = [{ type = "effort", values = ["high"] }] temperature = true +[[reasoning_options]] +type = "effort" +values = ["high"] + [cost] input = 0.95 output = 4 diff --git a/providers/cloudflare-workers-ai/models/@cf/nvidia/nemotron-3-120b-a12b.toml b/providers/cloudflare-workers-ai/models/@cf/nvidia/nemotron-3-120b-a12b.toml index 554decfc2e3..a75a99e22b1 100644 --- a/providers/cloudflare-workers-ai/models/@cf/nvidia/nemotron-3-120b-a12b.toml +++ b/providers/cloudflare-workers-ai/models/@cf/nvidia/nemotron-3-120b-a12b.toml @@ -5,9 +5,15 @@ base_model = "nvidia/nemotron-3-super-120b-a12b" name = "Nemotron 3 Super 120B" structured_output = true -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "high"] }] interleaved = true +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "medium", "high"] + [cost] input = 0.5 output = 1.5 diff --git a/providers/cloudflare-workers-ai/models/@cf/qwen/qwen3.8-27b.toml b/providers/cloudflare-workers-ai/models/@cf/qwen/qwen3.8-27b.toml index 307aa38fb79..f491dd473a3 100644 --- a/providers/cloudflare-workers-ai/models/@cf/qwen/qwen3.8-27b.toml +++ b/providers/cloudflare-workers-ai/models/@cf/qwen/qwen3.8-27b.toml @@ -5,7 +5,13 @@ base_model = "alibaba/qwen3.8-27b" description = "Qwen vision-language model for visual reasoning, documents, and agent tasks" -reasoning_options = [{ type = "toggle" }, { type = "effort", values = ["low", "medium", "xhigh"] }] + +[[reasoning_options]] +type = "toggle" + +[[reasoning_options]] +type = "effort" +values = ["low", "medium", "xhigh"] [cost] input = 0.45 diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml index e2b003ca0af..954aa924284 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml @@ -1,6 +1,5 @@ -# Native `/ai/run` accepts `reasoning_effort = low|medium|high` and -# `chat_template_kwargs.enable_thinking = true|false`; no budget is documented. -# https://developers.cloudflare.com/workers-ai/models/glm-4.7-flash/sync-input.json (accessed 2026-06-25) +# Toggle: chat_template_kwargs.enable_thinking = true|false (default: true). +# Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "zhipuai/glm-4.7-flash" description = "Efficient GLM model for fast reasoning, coding, and agent workflows" structured_output = true @@ -11,10 +10,6 @@ field = "reasoning_content" [[reasoning_options]] type = "toggle" -[[reasoning_options]] -type = "effort" -values = ["low", "medium", "high"] - [cost] input = 0.0605 output = 0.4 diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml index 968bc3c9482..60cbebe576e 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml @@ -3,7 +3,10 @@ base_model = "zhipuai/glm-5.3-flash" name = "Glm 5.3 Flash" description = "GLM vision model for visual reasoning, documents, and multimodal agents" -reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] [cost] input = 0.15 diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml index 52edd473a26..434f2b71106 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml @@ -3,7 +3,10 @@ base_model = "zhipuai/glm-5.3" name = "Glm 5.3" description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" -reasoning_options = [{ type = "effort", values = ["low", "high", "max"] }] + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] [cost] input = 1.4 From 767498eef07d56328d84333df57c4512afe99d47 Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 13:42:45 -0700 Subject: [PATCH 3/7] fix(cloudflare): cap oversized context and output limits --- .../models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml | 4 +++- .../models/@cf/zai-org/glm-5.3-flash.toml | 4 +++- .../cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml | 6 ++++-- 3 files changed, 10 insertions(+), 4 deletions(-) diff --git a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml index 84b4fc57fd4..a19afda0e78 100644 --- a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml +++ b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = none|low|high|max; none disables thinking. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Context/output capped to Cloudflare Workers AI context window (1,048,576). +# Search reports 1,310,720, exceeding AI Gateway max_completion_tokens. base_model = "deepseek/deepseek-v4-flash-0731" [[reasoning_options]] @@ -12,5 +14,5 @@ output = 1.32 cache_read = 0.014 [limit] -context = 1_310_720 +context = 1_048_576 output = 1_048_576 diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml index 60cbebe576e..83a85d6c195 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3-flash.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = low|high|max; thinking is mandatory. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Context/output capped to Cloudflare Workers AI context window (1,048,576). +# Search reports 1,310,720, exceeding AI Gateway max_completion_tokens. base_model = "zhipuai/glm-5.3-flash" name = "Glm 5.3 Flash" description = "GLM vision model for visual reasoning, documents, and multimodal agents" @@ -14,7 +16,7 @@ output = 0.5 cache_read = 0.03 [limit] -context = 1_310_720 +context = 1_048_576 output = 1_048_576 [modalities] diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml index 434f2b71106..a66154848f3 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-5.3.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = low|high|max; thinking is mandatory. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Context/output capped to Cloudflare Workers AI context window (1,048,576). +# Search reports 1,310,720, exceeding AI Gateway max_completion_tokens. base_model = "zhipuai/glm-5.3" name = "Glm 5.3" description = "Flagship GLM model for hybrid reasoning, coding, and agentic engineering" @@ -14,5 +16,5 @@ output = 4.4 cache_read = 0.26 [limit] -context = 1_310_720 -output = 1_310_720 +context = 1_048_576 +output = 1_048_576 From e892618007f94cb6b29eb8c1e7deda06c8559857 Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 16:45:43 -0700 Subject: [PATCH 4/7] fix(cloudflare): inherit SEA-LION description and clarify reasoning sources --- .../models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml | 1 - .../models/@cf/google/gemma-4-26b-a4b-it.toml | 2 ++ .../cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml | 2 ++ .../cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml | 2 ++ 4 files changed, 6 insertions(+), 1 deletion(-) diff --git a/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml b/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml index 8e5628d902d..3fe191a4d99 100644 --- a/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml +++ b/providers/cloudflare-workers-ai/models/@cf/aisingapore/gemma-sea-lion-v4-27b-it.toml @@ -1,6 +1,5 @@ base_model = "aisingapore/gemma-sea-lion-v4-27b-it" name = "Gemma Sea Lion V4 27B It" -description = "Open Gemma instruction model for efficient chat and self-hosted deployments" structured_output = false [cost] diff --git a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml index d2790a7f64a..2c6a8e4386f 100644 --- a/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml +++ b/providers/cloudflare-workers-ai/models/@cf/google/gemma-4-26b-a4b-it.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = none|high; none disables thinking. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Per-model Search config normalizes low/medium to high; these are aliases, +# not distinct levels despite the generic sync-input.json effort enum. base_model = "google/gemma-4-26b-a4b-it" interleaved = true diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml index d3f174f1a8e..6148d7218ca 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.6.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = none|high; none selects instant mode. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Per-model Search config normalizes low/medium to high; these are aliases, +# not distinct levels despite the generic sync-input.json effort enum. base_model = "moonshotai/kimi-k2.6" diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml index 954aa924284..9a09b252d06 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml @@ -1,5 +1,7 @@ # Toggle: chat_template_kwargs.enable_thinking = true|false (default: true). # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# Workers AI's per-model config exposes only the binary thinking toggle; +# the generic sync-input.json low/medium/high enum is not an effort selector here. base_model = "zhipuai/glm-4.7-flash" description = "Efficient GLM model for fast reasoning, coding, and agent workflows" structured_output = true From bc88758d6794aa1b139130af4b2f5a787323c62c Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 16:48:51 -0700 Subject: [PATCH 5/7] docs(cloudflare): cite GLM thinking control at its catalogue entry --- .../cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml | 1 + 1 file changed, 1 insertion(+) diff --git a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml index 9a09b252d06..37f0d8a203e 100644 --- a/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml +++ b/providers/cloudflare-workers-ai/models/@cf/zai-org/glm-4.7-flash.toml @@ -2,6 +2,7 @@ # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). # Workers AI's per-model config exposes only the binary thinking toggle; # the generic sync-input.json low/medium/high enum is not an effort selector here. +# Creator template: https://huggingface.co/zai-org/GLM-4.7-Flash/blob/main/chat_template.jinja base_model = "zhipuai/glm-4.7-flash" description = "Efficient GLM model for fast reasoning, coding, and agent workflows" structured_output = true From 0b12ebe1b2946fa37c5b2990413415656e76208c Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 17:12:09 -0700 Subject: [PATCH 6/7] docs(cloudflare): cite dated DeepSeek effort controls --- .../models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml | 2 ++ .../models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml | 2 ++ 2 files changed, 4 insertions(+) diff --git a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml index a19afda0e78..2385a1bc60b 100644 --- a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml +++ b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-flash-0731.toml @@ -2,6 +2,8 @@ # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). # Context/output capped to Cloudflare Workers AI context window (1,048,576). # Search reports 1,310,720, exceeding AI Gateway max_completion_tokens. +# This dated release adds a distinct low effort; the older preview does not. +# Creator: https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731#chat-template base_model = "deepseek/deepseek-v4-flash-0731" [[reasoning_options]] diff --git a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml index 262c61fe520..32c7c6708b0 100644 --- a/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml +++ b/providers/cloudflare-workers-ai/models/@cf/deepseek-ai/deepseek-v4-pro-0813.toml @@ -1,5 +1,7 @@ # Effort: reasoning_effort = none|low|high|max; none disables thinking. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). +# This dated release adds a distinct low effort; the older preview does not. +# Creator: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813#chat-template base_model = "deepseek/deepseek-v4-pro-0813" [[reasoning_options]] From 62e43e90a4c9687aab23eddd555d5c7f7a7c55d5 Mon Sep 17 00:00:00 2001 From: Kastan Day Date: Fri, 18 Sep 2026 17:59:22 -0700 Subject: [PATCH 7/7] fix(cloudflare): omit fixed Kimi K2.7 effort control --- .../models/@cf/moonshotai/kimi-k2.7-code.toml | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) diff --git a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml index 79685ac02e4..43a8509e3a6 100644 --- a/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml +++ b/providers/cloudflare-workers-ai/models/@cf/moonshotai/kimi-k2.7-code.toml @@ -1,11 +1,8 @@ -# Effort: reasoning_effort = high; thinking is mandatory. +# Thinking is mandatory; high is the only effective effort, so aliases add no caller control. # Source: Workers AI /ai/models/search?format=openrouter (2026-09-18). base_model = "moonshotai/kimi-k2.7-code" temperature = true - -[[reasoning_options]] -type = "effort" -values = ["high"] +reasoning_options = [] [cost] input = 0.95