diff --git a/models.yaml b/models.yaml index ebd92b4..a4c7fc8 100644 --- a/models.yaml +++ b/models.yaml @@ -863,9 +863,9 @@ - name: gemma-4-26b-a4b-it type: chat max_input_tokens: 262144 - input_price: 0.09 - output_price: 0.3 - cache_read_price: 0.05 + input_price: 0.0675 + output_price: 0.225 + cache_read_price: 0.0375 max_output_tokens: 32768 supports_vision: true supports_function_calling: true @@ -5764,7 +5764,6 @@ max_input_tokens: 262144 input_price: 0.6 output_price: 3.6 - cache_read_price: 0.15 max_output_tokens: 65536 supports_vision: true supports_function_calling: true @@ -7371,15 +7370,15 @@ - name: deepseek/deepseek-v4-flash type: chat max_input_tokens: 1048576 - input_price: 0.049 - output_price: 0.098 - cache_read_price: 0.0098 + input_price: 0.047 + output_price: 0.0941 + cache_read_price: 0.0094 max_output_tokens: 384000 supports_function_calling: true - name: deepseek/deepseek-v4-flash-0731 type: chat max_input_tokens: 1310720 - input_price: 0.03 + input_price: 0.022 output_price: 0.32 cache_read_price: 0.016 max_output_tokens: 943718 @@ -7396,26 +7395,26 @@ - name: deepseek/deepseek-v4-pro type: chat max_input_tokens: 1048576 - input_price: 0.783 - output_price: 1.566 - cache_read_price: 0.0653 + input_price: 0.4392 + output_price: 0.8784 + cache_read_price: 0.0366 max_output_tokens: 384000 supports_function_calling: true - name: deepseek/deepseek-v4-pro-0813 type: chat max_input_tokens: 1048576 - input_price: 0.462 - output_price: 1.386 - cache_read_price: 0.0154 + input_price: 0.264 + output_price: 0.792 + cache_read_price: 0.0088 max_output_tokens: 384000 supports_function_calling: true - name: deepseek/deepseek-v4.1-flash type: chat max_input_tokens: 1048576 - input_price: 0.15 - output_price: 0.6 - cache_read_price: 0.003 - max_output_tokens: 393216 + input_price: 0.14 + output_price: 0.42 + cache_read_price: 0.0042 + max_output_tokens: 131072 supports_vision: true supports_function_calling: true - name: deepseek/deepseek-v4.1-flash:batch @@ -7799,9 +7798,9 @@ - name: google/gemma-4-26b-a4b-it type: chat max_input_tokens: 262144 - input_price: 0.09 - output_price: 0.3 - cache_read_price: 0.05 + input_price: 0.0675 + output_price: 0.225 + cache_read_price: 0.0375 max_output_tokens: 235929 supports_vision: true supports_function_calling: true @@ -7913,9 +7912,9 @@ - name: inclusionai/ling-3.0-flash-vl type: chat max_input_tokens: 262144 - input_price: 0.06 - output_price: 0.18 - cache_read_price: 0.012 + input_price: 0.021 + output_price: 0.0616 + cache_read_price: 0.0042 max_output_tokens: 32768 supports_vision: true supports_function_calling: true @@ -8433,22 +8432,6 @@ input_price: 0.9 output_price: 1.9 max_output_tokens: 131072 - - name: nex-agi/nex-n2.5-mini:free - type: chat - max_input_tokens: 262144 - input_price: 0.0 - output_price: 0.0 - max_output_tokens: 235929 - supports_vision: true - supports_function_calling: true - - name: nex-agi/nex-n2.5-pro:free - type: chat - max_input_tokens: 262144 - input_price: 0.0 - output_price: 0.0 - max_output_tokens: 235929 - supports_vision: true - supports_function_calling: true - name: nousresearch/hermes-3-llama-3.1-405b type: chat max_input_tokens: 131072 @@ -8529,10 +8512,10 @@ - name: nvidia/nemotron-3.5-lightning type: chat max_input_tokens: 1000000 - input_price: 0.07 + input_price: 0.08 output_price: 0.2 cache_read_price: 0.04 - max_output_tokens: 235929 + max_output_tokens: 131072 supports_function_calling: true - name: nvidia/nemotron-3.5-lightning:free type: chat @@ -9607,6 +9590,14 @@ output_price: 1.5 max_output_tokens: 8192 supports_vision: true + - name: perceptron/perceptron-mk1.5 + type: chat + max_input_tokens: 36864 + input_price: 0.15 + output_price: 1.5 + max_output_tokens: 8192 + supports_vision: true + supports_function_calling: true - name: perplexity/sonar type: chat max_input_tokens: 127072 @@ -9746,9 +9737,9 @@ - name: qwen/qwen3-235b-a22b-2507 type: chat max_input_tokens: 262144 - input_price: 0.0875 - output_price: 0.35 - cache_read_price: 0.0175 + input_price: 0.35 + output_price: 1.4 + cache_read_price: 0.07 max_output_tokens: 235929 supports_function_calling: true - name: qwen/qwen3-235b-a22b-thinking-2507 @@ -9993,9 +9984,8 @@ type: chat max_input_tokens: 262144 input_price: 0.32 - output_price: 2.7 - cache_read_price: 0.15 - max_output_tokens: 262140 + output_price: 3.2 + max_output_tokens: 81920 supports_vision: true supports_function_calling: true - name: qwen/qwen3.6-35b-a3b @@ -10301,7 +10291,7 @@ max_output_tokens: 819200 - name: thinkingmachines/inkling type: chat - max_input_tokens: 1048576 + max_input_tokens: 524288 input_price: 1.0 output_price: 4.05 cache_read_price: 0.17 @@ -10310,7 +10300,7 @@ supports_function_calling: true - name: thinkingmachines/inkling-small type: chat - max_input_tokens: 1048576 + max_input_tokens: 524288 input_price: 0.45 output_price: 1.2 cache_read_price: 0.1 @@ -10333,6 +10323,10 @@ max_output_tokens: 262144 supports_vision: true supports_function_calling: true + - name: typesafe/jev-router + type: chat + max_input_tokens: 1000000 + supports_vision: true - name: unbiased/pareto type: chat max_input_tokens: 262144 @@ -10477,7 +10471,7 @@ supports_function_calling: true - name: xiaomi/mimo-v2.6-pro type: chat - max_input_tokens: 1048576 + max_input_tokens: 1050000 input_price: 0.435 output_price: 0.87 cache_read_price: 0.0036 @@ -10582,12 +10576,6 @@ cache_read_price: 0.1206 max_output_tokens: 131072 supports_function_calling: true - - name: z-ai/glm-5.2:free - type: chat - max_input_tokens: 32768 - input_price: 0.0 - output_price: 0.0 - max_output_tokens: 29491 - name: z-ai/glm-5.3 type: chat max_input_tokens: 1310720 @@ -10691,24 +10679,24 @@ - name: ~deepseek/deepseek-flash-latest type: chat max_input_tokens: 1048576 - input_price: 0.04 - output_price: 0.49 - cache_read_price: 0.01 - max_output_tokens: 943718 + input_price: 0.035 + output_price: 0.29 + cache_read_price: 0.001 + max_output_tokens: 384000 supports_vision: true supports_function_calling: true - name: ~deepseek/deepseek-pro-latest type: chat max_input_tokens: 1048576 - input_price: 0.2528 - output_price: 2.88 - cache_read_price: 0.0883 + input_price: 0.25 + output_price: 3.5 + cache_read_price: 0.249 max_output_tokens: 943718 supports_function_calling: true - name: ~deepseek/deepseek-v4-flash-latest type: chat max_input_tokens: 1310720 - input_price: 0.03 + input_price: 0.022 output_price: 0.32 cache_read_price: 0.016 max_output_tokens: 943718 @@ -10736,9 +10724,9 @@ - name: ~moonshotai/kimi-latest type: chat max_input_tokens: 1048576 - input_price: 0.8845 - output_price: 10.5346 - cache_read_price: 0.33 + input_price: 1.0301 + output_price: 9.043 + cache_read_price: 0.3 max_output_tokens: 943718 supports_vision: true supports_function_calling: true @@ -10812,10 +10800,10 @@ - name: ~z-ai/glm-latest type: chat max_input_tokens: 1310720 - input_price: 0.5614 - output_price: 1.7644 - cache_read_price: 0.1043 - max_output_tokens: 131072 + input_price: 0.5593 + output_price: 1.7578 + cache_read_price: 0.0919 + max_output_tokens: 943718 supports_function_calling: true - provider: github models: