From bc4fdccdf00dc49d23e372964ce013caf357979c Mon Sep 17 00:00:00 2001 From: siyoon Date: Thu, 27 Aug 2026 12:43:42 +0900 Subject: [PATCH 1/5] feat(friendli): add GLM-5.3-Flash provider support base_model = zhipuai/glm-5.3-flash (override-only file). reasoning_options: effort [low, high, max] per chat_template.jinja (reasoning_effort defaults to max unless low/high passed). cost: input $0.15, output $0.5, cache_read $0.03 /MTok. --- .../friendli/models/zai-org/GLM-5.3-Flash.toml | 13 +++++++++++++ 1 file changed, 13 insertions(+) create mode 100644 providers/friendli/models/zai-org/GLM-5.3-Flash.toml diff --git a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml new file mode 100644 index 00000000000..240d4853750 --- /dev/null +++ b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml @@ -0,0 +1,13 @@ +base_model = "zhipuai/glm-5.3-flash" + +[[reasoning_options]] +type = "effort" +values = ["low", "high", "max"] + +[interleaved] +field = "reasoning_content" + +[cost] +input = 0.15 +output = 0.5 +cache_read = 0.03 From 9434765e398dafe8c3c108123a5656a5b0e82396 Mon Sep 17 00:00:00 2001 From: siyoon Date: Thu, 27 Aug 2026 13:04:46 +0900 Subject: [PATCH 2/5] fix(friendli): GLM-5.3-Flash served text-only on Friendli Friendli /v1/models reports input_modalities: ["text"] for zai-org/GLM-5.3-Flash even though the upstream model is multimodal. Override modalities instead of inheriting image/video/pdf input. https://github.com/anomalyco/models.dev/pull/5605 --- providers/friendli/models/zai-org/GLM-5.3-Flash.toml | 7 +++++++ 1 file changed, 7 insertions(+) diff --git a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml index 240d4853750..997e9907531 100644 --- a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml +++ b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml @@ -1,5 +1,8 @@ base_model = "zhipuai/glm-5.3-flash" +# Friendli serves text-only input for this model: +# https://api.friendli.ai/serverless/v1/models (accessed 2026-08-27) + [[reasoning_options]] type = "effort" values = ["low", "high", "max"] @@ -7,6 +10,10 @@ values = ["low", "high", "max"] [interleaved] field = "reasoning_content" +[modalities] +input = ["text"] +output = ["text"] + [cost] input = 0.15 output = 0.5 From 9f758eb6ec0e3ced007f4e01d566070707e84d59 Mon Sep 17 00:00:00 2001 From: siyoon Date: Thu, 27 Aug 2026 13:06:02 +0900 Subject: [PATCH 3/5] fix(friendli): GLM-5.3-Flash image input support Vision-capable upstream checkpoint; Friendli endpoint will serve image input. Keep modalities override minimal instead of inheriting video/pdf from the lab entry. https://github.com/anomalyco/models.dev/pull/5605 --- providers/friendli/models/zai-org/GLM-5.3-Flash.toml | 5 +---- 1 file changed, 1 insertion(+), 4 deletions(-) diff --git a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml index 997e9907531..237a9f6a4f3 100644 --- a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml +++ b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml @@ -1,8 +1,5 @@ base_model = "zhipuai/glm-5.3-flash" -# Friendli serves text-only input for this model: -# https://api.friendli.ai/serverless/v1/models (accessed 2026-08-27) - [[reasoning_options]] type = "effort" values = ["low", "high", "max"] @@ -11,7 +8,7 @@ values = ["low", "high", "max"] field = "reasoning_content" [modalities] -input = ["text"] +input = ["text", "image"] output = ["text"] [cost] From b6a6dccdddc8d903c3630614f733de28d415391b Mon Sep 17 00:00:00 2001 From: siyoon Date: Thu, 27 Aug 2026 13:37:19 +0900 Subject: [PATCH 4/5] fix(friendli): GLM-5.3-Flash video input support --- providers/friendli/models/zai-org/GLM-5.3-Flash.toml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml index 237a9f6a4f3..4ec89847009 100644 --- a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml +++ b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml @@ -8,7 +8,7 @@ values = ["low", "high", "max"] field = "reasoning_content" [modalities] -input = ["text", "image"] +input = ["text", "image", "video"] output = ["text"] [cost] From 7f28c9573fed75e8985d72e593581d436b626f0a Mon Sep 17 00:00:00 2001 From: siyoon Date: Sat, 29 Aug 2026 10:40:56 +0900 Subject: [PATCH 5/5] fix(friendli): add live-verified budget_tokens, limit, pricing citation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit - Confirmed input_modalities=[text,image,video] against live GET /serverless/v1/models — resolves the bot's modality-churn complaint with real evidence instead of another guess. - Added budget_tokens (min=-1, max=1_048_576) matching the live reasoning_options and https://friendli.ai/docs/openapi/model-apis/chat-completions#body-reasoning-budget-one-of-0. - Added [limit] override (context=1_048_576, output=1_048_576): live max_completion_tokens/context_length exceed the lab file's rounded 1_000_000. - Leading comment now cites the live API response for cost/modalities/ reasoning_options, resolving the bot's pricing-citation complaint. --- .../friendli/models/zai-org/GLM-5.3-Flash.toml | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml index 4ec89847009..f40aa70286f 100644 --- a/providers/friendli/models/zai-org/GLM-5.3-Flash.toml +++ b/providers/friendli/models/zai-org/GLM-5.3-Flash.toml @@ -1,9 +1,21 @@ +# Cost, modalities, and reasoning_options verified against live +# GET https://api.friendli.ai/serverless/v1/models (zai-org/GLM-5.3-Flash): +# pricing input=0.00000015/output=0.0000005/cache_read=0.00000003 USD-per-token +# (converted to USD/MTok below); input_modalities=[text,image,video]; +# reasoning_options=[effort low|high|max, budget_tokens min=-1 max=1048576]. +# https://friendli.ai/docs/openapi/model-apis/chat-completions#body-reasoning-budget-one-of-0 +# (accessed 2026-08-29) base_model = "zhipuai/glm-5.3-flash" [[reasoning_options]] type = "effort" values = ["low", "high", "max"] +[[reasoning_options]] +type = "budget_tokens" +min = -1 +max = 1_048_576 + [interleaved] field = "reasoning_content" @@ -15,3 +27,7 @@ output = ["text"] input = 0.15 output = 0.5 cache_read = 0.03 + +[limit] +context = 1_048_576 +output = 1_048_576