diff --git a/providers/chutes/models/Alibaba-NLP/Tongyi-DeepResearch-30B-A3B.toml b/providers/chutes/models/Alibaba-NLP/Tongyi-DeepResearch-30B-A3B.toml deleted file mode 100644 index 18a36ec1..00000000 --- a/providers/chutes/models/Alibaba-NLP/Tongyi-DeepResearch-30B-A3B.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "Tongyi DeepResearch 30B A3B" -family = "alibaba-nlp" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = true -temperature = true -tool_call = true -structured_output = true -open_weights = true - -[cost] -input = 0.1 -output = 0.39 -reasoning = 0.585 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 131072 -output = 131072 - -[modalities] -input = ['text'] -output = ['text'] - -[interleaved] -field = "reasoning_content" diff --git a/providers/chutes/models/ArliAI/QwQ-32B-ArliAI-RpR-v1.toml b/providers/chutes/models/ArliAI/QwQ-32B-ArliAI-RpR-v1.toml deleted file mode 100644 index e901d0a6..00000000 --- a/providers/chutes/models/ArliAI/QwQ-32B-ArliAI-RpR-v1.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "QwQ 32B ArliAI RpR v1" -family = "arliai" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = true -temperature = true -tool_call = false -structured_output = true -open_weights = true - -[cost] -input = 0.03 -output = 0.11 -reasoning = 0.165 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 32768 -output = 32768 - -[modalities] -input = ['text'] -output = ['text'] - -[interleaved] -field = "reasoning_content" diff --git a/providers/chutes/models/MiniMaxAI/MiniMax-M2.1-TEE.toml b/providers/chutes/models/MiniMaxAI/MiniMax-M2.1-TEE.toml index 04f55fed..a3bd6934 100644 --- a/providers/chutes/models/MiniMaxAI/MiniMax-M2.1-TEE.toml +++ b/providers/chutes/models/MiniMaxAI/MiniMax-M2.1-TEE.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.3 output = 1.2 -reasoning = 1.7999999999999998 +reasoning = 1.8 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/MiniMaxAI/MiniMax-M2.toml b/providers/chutes/models/MiniMaxAI/MiniMax-M2.toml deleted file mode 100644 index 8bc40db8..00000000 --- a/providers/chutes/models/MiniMaxAI/MiniMax-M2.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "MiniMax M2" -family = "minimaxai" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = true -temperature = true -tool_call = true -structured_output = true -open_weights = true - -[cost] -input = 0.26 -output = 1.02 -reasoning = 1.53 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 196608 -output = 196608 - -[modalities] -input = ['text'] -output = ['text'] - -[interleaved] -field = "reasoning_content" diff --git a/providers/chutes/models/NousResearch/Hermes-4-14B.toml b/providers/chutes/models/NousResearch/Hermes-4-14B.toml index 843e3abc..8ef976ab 100644 --- a/providers/chutes/models/NousResearch/Hermes-4-14B.toml +++ b/providers/chutes/models/NousResearch/Hermes-4-14B.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.01 output = 0.05 -reasoning = 0.07500000000000001 +reasoning = 0.075 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/NousResearch/Hermes-4-405B-FP8-TEE.toml b/providers/chutes/models/NousResearch/Hermes-4-405B-FP8-TEE.toml index bc8818ba..b5ac43ba 100644 --- a/providers/chutes/models/NousResearch/Hermes-4-405B-FP8-TEE.toml +++ b/providers/chutes/models/NousResearch/Hermes-4-405B-FP8-TEE.toml @@ -1,26 +1,26 @@ name = "Hermes 4 405B FP8 TEE" family = "nousresearch" release_date = "2025-12-29" -last_updated = "2025-12-29" +last_updated = "2025-01-06" attachment = false -reasoning = false +reasoning = true temperature = true -tool_call = false -structured_output = false +tool_call = true +structured_output = true open_weights = true [cost] input = 0.3 output = 1.2 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 +reasoning = 1.8 [limit] -context = 40960 -output = 40960 +context = 131072 +output = 65536 [modalities] input = ['text'] output = ['text'] + +[interleaved] +field = "reasoning_content" diff --git a/providers/chutes/models/NousResearch/Hermes-4-405B-FP8.toml b/providers/chutes/models/NousResearch/Hermes-4-405B-FP8.toml deleted file mode 100644 index e0a5b66e..00000000 --- a/providers/chutes/models/NousResearch/Hermes-4-405B-FP8.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "Hermes 4 405B FP8" -family = "nousresearch" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = true -temperature = true -tool_call = true -structured_output = true -open_weights = true - -[cost] -input = 0.3 -output = 1.2 -reasoning = 1.7999999999999998 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 131072 -output = 131072 - -[modalities] -input = ['text'] -output = ['text'] - -[interleaved] -field = "reasoning_content" diff --git a/providers/chutes/models/NousResearch/Hermes-4-70B.toml b/providers/chutes/models/NousResearch/Hermes-4-70B.toml index 3dfab2a1..73d66ad1 100644 --- a/providers/chutes/models/NousResearch/Hermes-4-70B.toml +++ b/providers/chutes/models/NousResearch/Hermes-4-70B.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.11 output = 0.38 -reasoning = 0.5700000000000001 +reasoning = 0.57 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/NousResearch/Hermes-4.3-36B.toml b/providers/chutes/models/NousResearch/Hermes-4.3-36B.toml index f4cfcd40..580bc714 100644 --- a/providers/chutes/models/NousResearch/Hermes-4.3-36B.toml +++ b/providers/chutes/models/NousResearch/Hermes-4.3-36B.toml @@ -18,8 +18,8 @@ input_audio = 0.0 output_audio = 0.0 [limit] -context = 40960 -output = 40960 +context = 524288 +output = 524288 [modalities] input = ['text'] diff --git a/providers/chutes/models/OpenGVLab/InternVL3-78B.toml b/providers/chutes/models/OpenGVLab/InternVL3-78B-TEE.toml similarity index 64% rename from providers/chutes/models/OpenGVLab/InternVL3-78B.toml rename to providers/chutes/models/OpenGVLab/InternVL3-78B-TEE.toml index 4a40a9d0..78ef0952 100644 --- a/providers/chutes/models/OpenGVLab/InternVL3-78B.toml +++ b/providers/chutes/models/OpenGVLab/InternVL3-78B-TEE.toml @@ -1,7 +1,7 @@ -name = "InternVL3 78B" +name = "InternVL3 78B TEE" family = "opengvlab" -release_date = "2025-12-29" -last_updated = "2025-12-29" +release_date = "2025-01-06" +last_updated = "2025-01-06" attachment = false reasoning = false temperature = true @@ -12,10 +12,6 @@ open_weights = true [cost] input = 0.1 output = 0.39 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 [limit] context = 32768 @@ -24,3 +20,4 @@ output = 32768 [modalities] input = ['text', 'image'] output = ['text'] + diff --git a/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct-TEE.toml b/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct-TEE.toml index cc9f6ae7..054321af 100644 --- a/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct-TEE.toml +++ b/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct-TEE.toml @@ -22,5 +22,5 @@ context = 40960 output = 40960 [modalities] -input = ['text'] +input = ['text', 'image'] output = ['text'] diff --git a/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct.toml b/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct.toml deleted file mode 100644 index 176c8f8e..00000000 --- a/providers/chutes/models/Qwen/Qwen2.5-VL-72B-Instruct.toml +++ /dev/null @@ -1,26 +0,0 @@ -name = "Qwen2.5 VL 72B Instruct" -family = "qwen" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = false -temperature = true -tool_call = false -structured_output = true -open_weights = true - -[cost] -input = 0.07 -output = 0.26 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 32768 -output = 32768 - -[modalities] -input = ['text', 'image'] -output = ['text'] diff --git a/providers/chutes/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml b/providers/chutes/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml deleted file mode 100644 index 24db2a7f..00000000 --- a/providers/chutes/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml +++ /dev/null @@ -1,26 +0,0 @@ -name = "Qwen3 235B A22B Instruct 2507" -family = "qwen" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = false -temperature = true -tool_call = true -structured_output = true -open_weights = true - -[cost] -input = 0.08 -output = 0.55 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 262144 -output = 262144 - -[modalities] -input = ['text'] -output = ['text'] diff --git a/providers/chutes/models/Qwen/Qwen3-235B-A22B-Thinking-2507.toml b/providers/chutes/models/Qwen/Qwen3-235B-A22B-Thinking-2507.toml index f7cdb514..73b062b7 100644 --- a/providers/chutes/models/Qwen/Qwen3-235B-A22B-Thinking-2507.toml +++ b/providers/chutes/models/Qwen/Qwen3-235B-A22B-Thinking-2507.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.11 output = 0.6 -reasoning = 0.8999999999999999 +reasoning = 0.9 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/Qwen/Qwen3-235B-A22B.toml b/providers/chutes/models/Qwen/Qwen3-235B-A22B.toml index 67d0917f..94e8ce9b 100644 --- a/providers/chutes/models/Qwen/Qwen3-235B-A22B.toml +++ b/providers/chutes/models/Qwen/Qwen3-235B-A22B.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.3 output = 1.2 -reasoning = 1.7999999999999998 +reasoning = 1.8 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/Qwen/Qwen3-VL-235B-A22B-Thinking.toml b/providers/chutes/models/Qwen/Qwen3-VL-235B-A22B-Thinking.toml deleted file mode 100644 index f1452564..00000000 --- a/providers/chutes/models/Qwen/Qwen3-VL-235B-A22B-Thinking.toml +++ /dev/null @@ -1,30 +0,0 @@ -name = "Qwen3 VL 235B A22B Thinking" -family = "qwen" -release_date = "2025-12-29" -last_updated = "2025-12-29" -attachment = false -reasoning = true -temperature = true -tool_call = true -structured_output = true -open_weights = true - -[cost] -input = 0.3 -output = 1.2 -reasoning = 1.7999999999999998 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 - -[limit] -context = 262144 -output = 262144 - -[modalities] -input = ['text', 'image'] -output = ['text'] - -[interleaved] -field = "reasoning_content" diff --git a/providers/chutes/models/Qwen/Qwen3Guard-Gen-0.6B.toml b/providers/chutes/models/Qwen/Qwen3Guard-Gen-0.6B.toml index c6e379d2..dc304893 100644 --- a/providers/chutes/models/Qwen/Qwen3Guard-Gen-0.6B.toml +++ b/providers/chutes/models/Qwen/Qwen3Guard-Gen-0.6B.toml @@ -18,8 +18,8 @@ input_audio = 0.0 output_audio = 0.0 [limit] -context = 40960 -output = 40960 +context = 32768 +output = 32768 [modalities] input = ['text'] diff --git a/providers/chutes/models/XiaomiMiMo/MiMo-V2-Flash.toml b/providers/chutes/models/XiaomiMiMo/MiMo-V2-Flash.toml index b09c1e55..00ebb09e 100644 --- a/providers/chutes/models/XiaomiMiMo/MiMo-V2-Flash.toml +++ b/providers/chutes/models/XiaomiMiMo/MiMo-V2-Flash.toml @@ -18,8 +18,8 @@ input_audio = 0.0 output_audio = 0.0 [limit] -context = 40960 -output = 40960 +context = 262144 +output = 262144 [modalities] input = ['text'] diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-R1-0528-TEE.toml b/providers/chutes/models/deepseek-ai/DeepSeek-R1-0528-TEE.toml index dda96eb9..40755b82 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-R1-0528-TEE.toml +++ b/providers/chutes/models/deepseek-ai/DeepSeek-R1-0528-TEE.toml @@ -20,7 +20,7 @@ output_audio = 0.0 [limit] context = 163840 -output = 163840 +output = 65536 [modalities] input = ['text'] diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-R1-TEE.toml b/providers/chutes/models/deepseek-ai/DeepSeek-R1-TEE.toml index 7757d279..af3963c2 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-R1-TEE.toml +++ b/providers/chutes/models/deepseek-ai/DeepSeek-R1-TEE.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.3 output = 1.2 -reasoning = 1.7999999999999998 +reasoning = 1.8 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-V3-0324-TEE.toml b/providers/chutes/models/deepseek-ai/DeepSeek-V3-0324-TEE.toml index c4bb2c23..537c1723 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-V3-0324-TEE.toml +++ b/providers/chutes/models/deepseek-ai/DeepSeek-V3-0324-TEE.toml @@ -10,8 +10,8 @@ structured_output = true open_weights = true [cost] -input = 0.24 -output = 0.84 +input = 0.19 +output = 0.87 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-V3.1-TEE.toml b/providers/chutes/models/deepseek-ai/DeepSeek-V3.1-TEE.toml index d9ce2db4..1b026fe2 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-V3.1-TEE.toml +++ b/providers/chutes/models/deepseek-ai/DeepSeek-V3.1-TEE.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.2 output = 0.8 -reasoning = 1.2000000000000002 +reasoning = 1.2 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-V3.2-TEE.toml b/providers/chutes/models/deepseek-ai/DeepSeek-V3.2-TEE.toml index 9d66c45d..159b5d35 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-V3.2-TEE.toml +++ b/providers/chutes/models/deepseek-ai/DeepSeek-V3.2-TEE.toml @@ -10,9 +10,9 @@ structured_output = true open_weights = true [cost] -input = 0.27 -output = 0.41 -reasoning = 0.615 +input = 0.25 +output = 0.38 +reasoning = 0.57 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 @@ -20,7 +20,7 @@ output_audio = 0.0 [limit] context = 163840 -output = 16384 +output = 65536 [modalities] input = ['text'] diff --git a/providers/chutes/models/openai/gpt-oss-120b-TEE.toml b/providers/chutes/models/openai/gpt-oss-120b-TEE.toml index 9b900a4f..c3931143 100644 --- a/providers/chutes/models/openai/gpt-oss-120b-TEE.toml +++ b/providers/chutes/models/openai/gpt-oss-120b-TEE.toml @@ -11,8 +11,8 @@ open_weights = true [cost] input = 0.04 -output = 0.25 -reasoning = 0.375 +output = 0.18 +reasoning = 0.27 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/openai/gpt-oss-20b.toml b/providers/chutes/models/openai/gpt-oss-20b.toml index 905e04d4..401649bd 100644 --- a/providers/chutes/models/openai/gpt-oss-20b.toml +++ b/providers/chutes/models/openai/gpt-oss-20b.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.02 output = 0.1 -reasoning = 0.15000000000000002 +reasoning = 0.15 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/tngtech/DeepSeek-R1T-Chimera.toml b/providers/chutes/models/tngtech/DeepSeek-R1T-Chimera.toml index 4f69df1e..bf6d0103 100644 --- a/providers/chutes/models/tngtech/DeepSeek-R1T-Chimera.toml +++ b/providers/chutes/models/tngtech/DeepSeek-R1T-Chimera.toml @@ -12,7 +12,7 @@ open_weights = true [cost] input = 0.3 output = 1.2 -reasoning = 1.7999999999999998 +reasoning = 1.8 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/deepseek-ai/DeepSeek-V3.1.toml b/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera-TEE.toml similarity index 52% rename from providers/chutes/models/deepseek-ai/DeepSeek-V3.1.toml rename to providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera-TEE.toml index 78122acc..9368d301 100644 --- a/providers/chutes/models/deepseek-ai/DeepSeek-V3.1.toml +++ b/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera-TEE.toml @@ -1,7 +1,7 @@ -name = "DeepSeek V3.1" -family = "deepseek-ai" -release_date = "2025-12-29" -last_updated = "2025-12-29" +name = "DeepSeek TNG R1T2 Chimera TEE" +family = "tngtech" +release_date = "2025-01-06" +last_updated = "2025-01-06" attachment = false reasoning = true temperature = true @@ -10,13 +10,9 @@ structured_output = true open_weights = true [cost] -input = 0.2 -output = 0.8 -reasoning = 1.2000000000000002 -cache_read = 0.0 -cache_write = 0.0 -input_audio = 0.0 -output_audio = 0.0 +input = 0.25 +output = 0.85 +reasoning = 1.275 [limit] context = 163840 @@ -28,3 +24,4 @@ output = ['text'] [interleaved] field = "reasoning_content" + diff --git a/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera.toml b/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera.toml index 90c39839..bcedee94 100644 --- a/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera.toml +++ b/providers/chutes/models/tngtech/DeepSeek-TNG-R1T2-Chimera.toml @@ -10,9 +10,9 @@ structured_output = true open_weights = true [cost] -input = 0.3 -output = 1.2 -reasoning = 1.7999999999999998 +input = 0.25 +output = 0.85 +reasoning = 1.275 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/tngtech/TNG-R1T-Chimera-TEE.toml b/providers/chutes/models/tngtech/TNG-R1T-Chimera-TEE.toml index 914bd7d9..62d2fbee 100644 --- a/providers/chutes/models/tngtech/TNG-R1T-Chimera-TEE.toml +++ b/providers/chutes/models/tngtech/TNG-R1T-Chimera-TEE.toml @@ -10,9 +10,9 @@ structured_output = true open_weights = true [cost] -input = 0.3 -output = 1.2 -reasoning = 1.7999999999999998 +input = 0.25 +output = 0.85 +reasoning = 1.275 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0 diff --git a/providers/chutes/models/zai-org/GLM-4.6-TEE.toml b/providers/chutes/models/zai-org/GLM-4.6-TEE.toml index a7d08d9e..73a30219 100644 --- a/providers/chutes/models/zai-org/GLM-4.6-TEE.toml +++ b/providers/chutes/models/zai-org/GLM-4.6-TEE.toml @@ -10,9 +10,9 @@ structured_output = true open_weights = true [cost] -input = 0.4 -output = 1.75 -reasoning = 2.625 +input = 0.35 +output = 1.5 +reasoning = 2.25 cache_read = 0.0 cache_write = 0.0 input_audio = 0.0