diff --git a/models.yaml b/models.yaml index a1a68ef..d57ceb9 100644 --- a/models.yaml +++ b/models.yaml @@ -3567,13 +3567,6 @@ - medium - high - max - - name: mistral.devstral-2-123b - type: chat - max_input_tokens: 256000 - input_price: 0.4 - output_price: 2.0 - max_output_tokens: 8192 - supports_function_calling: true - name: us.anthropic.claude-sonnet-4-6 type: chat max_input_tokens: 1000000 @@ -3751,6 +3744,13 @@ output_price: 0.24 max_output_tokens: 8192 supports_function_calling: true + - name: mistral.devstral-2-123b + type: chat + max_input_tokens: 256000 + input_price: 0.4 + output_price: 2.0 + max_output_tokens: 8192 + supports_function_calling: true - name: amazon.nova-2-lite-v1:0 type: chat max_input_tokens: 1000000 @@ -3799,12 +3799,12 @@ - low - medium - high - - name: mistral.magistral-small-2509 + - name: mistral.ministral-3-14b-instruct type: chat max_input_tokens: 128000 - input_price: 0.5 - output_price: 1.5 - max_output_tokens: 40000 + input_price: 0.2 + output_price: 0.2 + max_output_tokens: 4096 supports_vision: true supports_function_calling: true - name: mistral.ministral-3-3b-instruct @@ -3815,6 +3815,14 @@ max_output_tokens: 8192 supports_vision: true supports_function_calling: true + - name: mistral.ministral-3-8b-instruct + type: chat + max_input_tokens: 128000 + input_price: 0.15 + output_price: 0.15 + max_output_tokens: 4096 + supports_vision: true + supports_function_calling: true - name: mistral.mistral-large-3-675b-instruct type: chat max_input_tokens: 256000 @@ -4095,6 +4103,14 @@ max_output_tokens: 262000 supports_vision: true supports_function_calling: true + - name: mistral.magistral-small-2509 + type: chat + max_input_tokens: 128000 + input_price: 0.5 + output_price: 1.5 + max_output_tokens: 40000 + supports_vision: true + supports_function_calling: true - name: qwen.qwen3-next-80b-a3b type: chat max_input_tokens: 262144 @@ -4220,11 +4236,18 @@ output_price: 0.88 max_output_tokens: 131072 supports_function_calling: true + - name: mistral.voxtral-mini-3b-2507 + type: chat + max_input_tokens: 32768 + input_price: 0.04 + output_price: 0.04 + max_output_tokens: 4096 + supports_function_calling: true - name: mistral.voxtral-small-24b-2507 type: chat - max_input_tokens: 32000 - input_price: 0.15 - output_price: 0.35 + max_input_tokens: 32768 + input_price: 0.1 + output_price: 0.3 max_output_tokens: 8192 supports_function_calling: true - name: us.amazon.nova-premier-v1:0 @@ -4235,13 +4258,6 @@ max_output_tokens: 5120 supports_vision: true supports_function_calling: true - - name: us.writer.palmyra-x4-v1:0 - type: chat - max_input_tokens: 122880 - input_price: 2.5 - output_price: 10.0 - max_output_tokens: 8192 - supports_function_calling: true - name: us.writer.palmyra-x5-v1:0 type: chat max_input_tokens: 1040000 @@ -4249,13 +4265,6 @@ output_price: 6.0 max_output_tokens: 8192 supports_function_calling: true - - name: writer.palmyra-x4-v1:0 - type: chat - max_input_tokens: 122880 - input_price: 2.5 - output_price: 10.0 - max_output_tokens: 8192 - supports_function_calling: true - name: writer.palmyra-x5-v1:0 type: chat max_input_tokens: 1040000 @@ -4292,15 +4301,15 @@ max_input_tokens: 1000000 input_price: 0.24 output_price: 0.97 - max_output_tokens: 16384 + max_output_tokens: 8192 supports_vision: true supports_function_calling: true - name: meta.llama4-scout-17b-instruct-v1:0 type: chat - max_input_tokens: 3500000 + max_input_tokens: 10000000 input_price: 0.17 output_price: 0.66 - max_output_tokens: 16384 + max_output_tokens: 8192 supports_vision: true supports_function_calling: true - name: us.meta.llama4-maverick-17b-instruct-v1:0 @@ -4308,16 +4317,16 @@ max_input_tokens: 1000000 input_price: 0.24 output_price: 0.97 - max_output_tokens: 16384 + max_output_tokens: 8192 require_max_tokens: true supports_vision: true supports_function_calling: true - name: us.meta.llama4-scout-17b-instruct-v1:0 type: chat - max_input_tokens: 3500000 + max_input_tokens: 10000000 input_price: 0.17 output_price: 0.66 - max_output_tokens: 16384 + max_output_tokens: 8192 require_max_tokens: true supports_vision: true supports_function_calling: true @@ -4476,26 +4485,19 @@ max_output_tokens: 5120 supports_vision: true supports_function_calling: true - - name: mistral.ministral-3-14b-instruct + - name: us.writer.palmyra-x4-v1:0 type: chat - max_input_tokens: 128000 - input_price: 0.2 - output_price: 0.2 - max_output_tokens: 4096 + max_input_tokens: 122880 + input_price: 2.5 + output_price: 10.0 + max_output_tokens: 8192 supports_function_calling: true - - name: mistral.ministral-3-8b-instruct + - name: writer.palmyra-x4-v1:0 type: chat - max_input_tokens: 128000 - input_price: 0.15 - output_price: 0.15 - max_output_tokens: 4096 - supports_function_calling: true - - name: mistral.voxtral-mini-3b-2507 - type: chat - max_input_tokens: 128000 - input_price: 0.04 - output_price: 0.04 - max_output_tokens: 4096 + max_input_tokens: 122880 + input_price: 2.5 + output_price: 10.0 + max_output_tokens: 8192 supports_function_calling: true - name: meta.llama3-1-70b-instruct-v1:0 type: chat @@ -6348,15 +6350,15 @@ - name: deepseek/deepseek-v4-flash type: chat max_input_tokens: 1048576 - input_price: 0.0676 - output_price: 0.1352 + input_price: 0.0668 + output_price: 0.1336 max_output_tokens: 384000 supports_function_calling: true - name: deepseek/deepseek-v4-flash-0731 type: chat max_input_tokens: 1310720 - input_price: 0.065 - output_price: 0.18 + input_price: 0.04 + output_price: 0.08 max_output_tokens: 943718 supports_function_calling: true - name: deepseek/deepseek-v4-flash-0731:batch @@ -6385,8 +6387,8 @@ - name: deepseek/deepseek-v4-pro type: chat max_input_tokens: 1048576 - input_price: 0.8397 - output_price: 1.6794 + input_price: 0.7788 + output_price: 1.5576 max_output_tokens: 384000 supports_function_calling: true - name: deepseek/deepseek-v4-pro-0813 @@ -6863,6 +6865,18 @@ max_output_tokens: 32768 supports_vision: true supports_function_calling: true + - name: inference-net/schematron-v2-small + type: chat + max_input_tokens: 128000 + input_price: 0.05 + output_price: 0.23 + max_output_tokens: 4096 + - name: inference-net/schematron-v2-turbo + type: chat + max_input_tokens: 128000 + input_price: 0.03 + output_price: 0.15 + max_output_tokens: 8192 - name: kwaipilot/kat-coder-pro-v2 type: chat max_input_tokens: 262144 @@ -7058,9 +7072,9 @@ - name: minimax/minimax-m2.5 type: chat max_input_tokens: 204800 - input_price: 0.3 - output_price: 1.2 - max_output_tokens: 131072 + input_price: 0.27 + output_price: 1.08 + max_output_tokens: 128000 supports_function_calling: true - name: minimax/minimax-m2.7 type: chat @@ -7292,7 +7306,7 @@ max_input_tokens: 262144 input_price: 0.6 output_price: 2.5 - max_output_tokens: 100352 + max_output_tokens: 235929 supports_function_calling: true - name: moonshotai/kimi-k2.5 type: chat @@ -7321,8 +7335,8 @@ - name: moonshotai/kimi-k3 type: chat max_input_tokens: 1048576 - input_price: 1.5352 - output_price: 7.7001 + input_price: 2.3027 + output_price: 11.5502 max_output_tokens: 943718 supports_vision: true supports_function_calling: true @@ -8988,7 +9002,7 @@ max_input_tokens: 1048576 input_price: 1.0 output_price: 4.05 - max_output_tokens: 32768 + max_output_tokens: 471859 supports_vision: true supports_function_calling: true - name: thinkingmachines/inkling-small @@ -9223,8 +9237,8 @@ - name: z-ai/glm-5.3-flash type: chat max_input_tokens: 1310720 - input_price: 0.15 - output_price: 0.5 + input_price: 0.075 + output_price: 0.25 max_output_tokens: 131072 supports_vision: true supports_function_calling: true @@ -9286,8 +9300,8 @@ - name: ~deepseek/deepseek-v4-flash-latest type: chat max_input_tokens: 1310720 - input_price: 0.05 - output_price: 0.16 + input_price: 0.03 + output_price: 0.07 max_output_tokens: 393216 supports_function_calling: true - name: ~google/gemini-flash-latest @@ -9309,8 +9323,8 @@ - name: ~moonshotai/kimi-latest type: chat max_input_tokens: 1048576 - input_price: 1.5352 - output_price: 7.7001 + input_price: 2.125 + output_price: 11.9 max_output_tokens: 943718 supports_vision: true supports_function_calling: true