diff --git a/models.yaml b/models.yaml index dfc8b22..88fbc8f 100644 --- a/models.yaml +++ b/models.yaml @@ -2681,6 +2681,48 @@ supports_function_calling: true - provider: bedrock models: + - name: global.openai.gpt-6-astra + type: chat + max_input_tokens: 1050000 + input_price: 10.0 + output_price: 50.0 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - xhigh + - max + - name: openai.gpt-6-astra + type: chat + max_input_tokens: 1050000 + input_price: 11.0 + output_price: 55.0 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - xhigh + - max + - name: us.openai.gpt-6-astra + type: chat + max_input_tokens: 1050000 + input_price: 11.0 + output_price: 55.0 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - xhigh + - max - name: anthropic.claude-fable-5-1 type: chat max_input_tokens: 1000000 @@ -2940,6 +2982,51 @@ - high - xhigh - max + - name: us.openai.gpt-5.6-luna + type: chat + max_input_tokens: 1050000 + input_price: 0.22 + output_price: 1.32 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - none + - low + - medium + - high + - xhigh + - max + - name: us.openai.gpt-5.6-sol + type: chat + max_input_tokens: 1050000 + input_price: 4.4 + output_price: 22.0 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - none + - low + - medium + - high + - xhigh + - max + - name: us.openai.gpt-5.6-terra + type: chat + max_input_tokens: 1050000 + input_price: 2.2 + output_price: 13.2 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - none + - low + - medium + - high + - xhigh + - max - name: anthropic.claude-sonnet-5 type: chat max_input_tokens: 1000000 @@ -3211,6 +3298,20 @@ - high - xhigh - max + - name: au.anthropic.claude-opus-4-7 + type: chat + max_input_tokens: 1000000 + input_price: 5.5 + output_price: 27.5 + max_output_tokens: 128000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - xhigh + - max - name: eu.anthropic.claude-opus-4-7 type: chat max_input_tokens: 1000000 @@ -3577,6 +3678,54 @@ output_price: 2.5 max_output_tokens: 16000 supports_function_calling: true + - name: eu.amazon.nova-2-lite-v1:0 + type: chat + max_input_tokens: 1000000 + input_price: 0.374 + output_price: 3.157 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - name: global.amazon.nova-2-lite-v1:0 + type: chat + max_input_tokens: 1000000 + input_price: 0.3 + output_price: 2.5 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - name: jp.amazon.nova-2-lite-v1:0 + type: chat + max_input_tokens: 1000000 + input_price: 0.396 + output_price: 3.311 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high + - name: us.amazon.nova-2-lite-v1:0 + type: chat + max_input_tokens: 1000000 + input_price: 0.33 + output_price: 2.75 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + reasoning_levels: + - low + - medium + - high - name: anthropic.claude-opus-4-5-20251101-v1:0 type: chat max_input_tokens: 200000 @@ -3925,6 +4074,20 @@ output_price: 0.35 max_output_tokens: 8192 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 + input_price: 0.6 + output_price: 6.0 + max_output_tokens: 8192 + supports_function_calling: true - name: writer.palmyra-x4-v1:0 type: chat max_input_tokens: 122880 @@ -3939,6 +4102,14 @@ output_price: 6.0 max_output_tokens: 8192 supports_function_calling: true + - name: eu.mistral.pixtral-large-2502-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 2.0 + output_price: 6.0 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true - name: mistral.pixtral-large-2502-v1:0 type: chat max_input_tokens: 128000 @@ -3947,6 +4118,14 @@ max_output_tokens: 8192 supports_vision: true supports_function_calling: true + - name: us.mistral.pixtral-large-2502-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 2.0 + output_price: 6.0 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true - name: meta.llama4-maverick-17b-instruct-v1:0 type: chat max_input_tokens: 1000000 @@ -4002,6 +4181,14 @@ output_price: 0.72 max_output_tokens: 4096 supports_function_calling: true + - name: us.meta.llama3-3-70b-instruct-v1:0 + type: chat + max_input_tokens: 131072 + input_price: 0.72 + output_price: 0.72 + max_output_tokens: 8192 + require_max_tokens: true + supports_function_calling: true - name: amazon.nova-lite-v1:0 type: chat max_input_tokens: 300000 @@ -4025,6 +4212,83 @@ max_output_tokens: 8192 supports_vision: true supports_function_calling: true + - name: apac.amazon.nova-lite-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.063 + output_price: 0.252 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true + - name: apac.amazon.nova-micro-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 0.037 + output_price: 0.148 + max_output_tokens: 8192 + supports_function_calling: true + - name: apac.amazon.nova-pro-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.84 + output_price: 3.36 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true + - name: ca.amazon.nova-lite-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.064 + output_price: 0.256 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true + - name: eu.amazon.nova-lite-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.069 + output_price: 0.276 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true + - name: eu.amazon.nova-micro-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 0.04 + output_price: 0.16 + max_output_tokens: 8192 + supports_function_calling: true + - name: eu.amazon.nova-pro-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.92 + output_price: 3.68 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true + - name: us.amazon.nova-lite-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.06 + output_price: 0.24 + max_output_tokens: 5120 + supports_vision: true + supports_function_calling: true + - name: us.amazon.nova-micro-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 0.035 + output_price: 0.14 + max_output_tokens: 5120 + supports_function_calling: true + - name: us.amazon.nova-pro-v1:0 + type: chat + max_input_tokens: 300000 + input_price: 0.8 + output_price: 3.2 + max_output_tokens: 5120 + supports_vision: true + supports_function_calling: true - name: amazon.nova-2-lite-v1:0 type: chat max_input_tokens: 128000 @@ -4102,6 +4366,22 @@ output_price: 0.22 max_output_tokens: 4096 supports_function_calling: true + - name: us.meta.llama3-1-70b-instruct-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 0.72 + output_price: 0.72 + max_output_tokens: 4096 + require_max_tokens: true + supports_function_calling: true + - name: us.meta.llama3-1-8b-instruct-v1:0 + type: chat + max_input_tokens: 128000 + input_price: 0.22 + output_price: 0.22 + max_output_tokens: 4096 + require_max_tokens: true + supports_function_calling: true - name: cohere.embed-english-v3 type: embedding input_price: 0.1 @@ -4114,40 +4394,12 @@ max_tokens_per_chunk: 512 default_chunk_size: 1000 max_batch_size: 96 - - name: us.meta.llama3-3-70b-instruct-v1:0 - type: chat - max_input_tokens: 131072 - input_price: 0.72 - output_price: 0.72 - max_output_tokens: 8192 - require_max_tokens: true - supports_function_calling: true - name: us.amazon.nova-premier-v1:0 type: chat max_input_tokens: 300000 input_price: 2.5 output_price: 12.5 max_output_tokens: 5120 - - name: us.amazon.nova-pro-v1:0 - type: chat - max_input_tokens: 300000 - input_price: 0.8 - output_price: 3.2 - max_output_tokens: 5120 - supports_vision: true - - name: us.amazon.nova-lite-v1:0 - type: chat - max_input_tokens: 300000 - input_price: 0.06 - output_price: 0.24 - max_output_tokens: 5120 - supports_vision: true - - name: us.amazon.nova-micro-v1:0 - type: chat - max_input_tokens: 128000 - input_price: 0.035 - output_price: 0.14 - max_output_tokens: 5120 - provider: cloudflare models: - name: '@cf/zai-org/glm-5.3-flash' @@ -5958,8 +6210,8 @@ - name: deepseek/deepseek-v4-flash-0731:batch type: chat max_input_tokens: 1048576 - input_price: 0.14 - output_price: 0.28 + input_price: 0.11 + output_price: 0.33 max_output_tokens: 943718 supports_function_calling: true - name: deepseek/deepseek-v4-flash-vision-exp @@ -5967,7 +6219,15 @@ max_input_tokens: 1048576 input_price: 0.22 output_price: 0.66 - max_output_tokens: 384000 + max_output_tokens: 943718 + supports_vision: true + supports_function_calling: true + - name: deepseek/deepseek-v4-flash-vision-exp:batch + type: chat + max_input_tokens: 1048576 + input_price: 0.11 + output_price: 0.33 + max_output_tokens: 943718 supports_vision: true supports_function_calling: true - name: deepseek/deepseek-v4-pro @@ -5987,8 +6247,8 @@ - name: deepseek/deepseek-v4-pro-0813:batch type: chat max_input_tokens: 1048576 - input_price: 1.32 - output_price: 3.96 + input_price: 0.66 + output_price: 1.98 max_output_tokens: 943718 supports_function_calling: true - name: dots-studio/dots-3-note-preview:free @@ -6381,8 +6641,8 @@ - name: ibm-granite/granite-4.2-8b type: chat max_input_tokens: 131072 - input_price: 0.1 - output_price: 0.15 + input_price: 0.06 + output_price: 0.25 max_output_tokens: 117964 supports_function_calling: true - name: inception/mercury-2 @@ -6392,7 +6652,7 @@ output_price: 0.75 max_output_tokens: 50000 supports_function_calling: true - - name: inception/mercury-2.5-preview + - name: inception/mercury-2.5 type: chat max_input_tokens: 260000 input_price: 0.04 @@ -6528,8 +6788,8 @@ - name: meta/muse-glimmer-30b:batch type: chat max_input_tokens: 131072 - input_price: 0.35 - output_price: 1.5 + input_price: 0.175 + output_price: 0.75 max_output_tokens: 117964 supports_vision: true supports_function_calling: true @@ -6595,7 +6855,7 @@ - name: minimax/minimax-m1 type: chat max_input_tokens: 1000000 - input_price: 0.4 + input_price: 0.55 output_price: 2.2 max_output_tokens: 40000 supports_function_calling: true @@ -6871,19 +7131,18 @@ input_price: 0.9 output_price: 1.9 max_output_tokens: 131072 - - name: nex-agi/nex-n2-mini + - name: nex-agi/nex-n2.5-mini:free type: chat max_input_tokens: 262144 - input_price: 0.025 - output_price: 0.1 + input_price: 0.0 + output_price: 0.0 max_output_tokens: 235929 - supports_vision: true supports_function_calling: true - - name: nex-agi/nex-n2-pro + - name: nex-agi/nex-n2.5-pro:free type: chat max_input_tokens: 262144 - input_price: 0.25 - output_price: 1.0 + input_price: 0.0 + output_price: 0.0 max_output_tokens: 235929 supports_vision: true supports_function_calling: true @@ -6928,7 +7187,7 @@ supports_function_calling: true - name: nvidia/nemotron-3-super-120b-a12b type: chat - max_input_tokens: 1000000 + max_input_tokens: 262144 input_price: 0.085 output_price: 0.4 max_output_tokens: 16384 @@ -8020,8 +8279,8 @@ - name: qwen/qwen3-235b-a22b-2507 type: chat max_input_tokens: 262144 - input_price: 0.09 - output_price: 0.55 + input_price: 0.22 + output_price: 0.88 max_output_tokens: 16384 supports_function_calling: true - name: qwen/qwen3-235b-a22b-thinking-2507 @@ -8118,9 +8377,9 @@ - name: qwen/qwen3-next-80b-a3b-instruct type: chat max_input_tokens: 262144 - input_price: 0.1 + input_price: 0.09 output_price: 1.1 - max_output_tokens: 235929 + max_output_tokens: 16384 supports_function_calling: true - name: qwen/qwen3-next-80b-a3b-thinking type: chat @@ -8324,7 +8583,7 @@ max_input_tokens: 1048576 input_price: 2.0 output_price: 6.0 - max_output_tokens: 262144 + max_output_tokens: 131072 supports_function_calling: true - name: qwen/qwen3.8-2.4t-a95b:batch type: chat @@ -8667,9 +8926,9 @@ - name: z-ai/glm-4.6 type: chat max_input_tokens: 204800 - input_price: 0.55 - output_price: 2.2 - max_output_tokens: 131072 + input_price: 0.43 + output_price: 1.75 + max_output_tokens: 16384 supports_function_calling: true - name: z-ai/glm-4.6v type: chat @@ -8721,6 +8980,13 @@ output_price: 3.036 max_output_tokens: 131072 supports_function_calling: true + - name: z-ai/glm-5.2:batch + type: chat + max_input_tokens: 1048576 + input_price: 0.7 + output_price: 2.2 + max_output_tokens: 943718 + supports_function_calling: true - name: z-ai/glm-5.3 type: chat max_input_tokens: 1310720 @@ -8738,12 +9004,19 @@ supports_function_calling: true - name: z-ai/glm-5.3-flash:batch type: chat - max_input_tokens: 1048575 - input_price: 0.15 - output_price: 0.5 - max_output_tokens: 943717 + max_input_tokens: 1048576 + input_price: 0.075 + output_price: 0.25 + max_output_tokens: 943718 supports_vision: true supports_function_calling: true + - name: z-ai/glm-5.3:batch + type: chat + max_input_tokens: 1048576 + input_price: 0.7 + output_price: 2.2 + max_output_tokens: 943718 + supports_function_calling: true - name: z-ai/glm-5v-turbo type: chat max_input_tokens: 202752 @@ -8810,8 +9083,8 @@ - name: ~moonshotai/kimi-latest type: chat max_input_tokens: 1048576 - input_price: 2.5 - output_price: 14.0 + input_price: 2.4 + output_price: 12.0 max_output_tokens: 943718 supports_vision: true supports_function_calling: true @@ -8842,17 +9115,17 @@ - name: ~z-ai/glm-flash-latest type: chat max_input_tokens: 1310720 - input_price: 0.0713 - output_price: 0.2375 - max_output_tokens: 131072 + input_price: 0.075 + output_price: 0.25 + max_output_tokens: 943718 supports_vision: true supports_function_calling: true - name: ~z-ai/glm-latest type: chat max_input_tokens: 1310720 - input_price: 1.12 - output_price: 3.52 - max_output_tokens: 943718 + input_price: 1.113 + output_price: 3.498 + max_output_tokens: 128000 supports_function_calling: true - provider: github models: