diff --git a/models.yaml b/models.yaml index 66ae8bb..85bcc29 100644 --- a/models.yaml +++ b/models.yaml @@ -3092,10 +3092,10 @@ - name: au.anthropic.claude-opus-5 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3140,10 +3140,10 @@ - name: jp.anthropic.claude-opus-5 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3156,10 +3156,10 @@ - name: us.anthropic.claude-opus-5 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -3377,10 +3377,10 @@ - name: au.anthropic.claude-sonnet-5 type: chat max_input_tokens: 1000000 - input_price: 2.0 - output_price: 10.0 - cache_read_price: 0.2 - cache_write_price: 2.5 + input_price: 2.2 + output_price: 11.0 + cache_read_price: 0.22 + cache_write_price: 2.75 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3425,10 +3425,10 @@ - name: jp.anthropic.claude-sonnet-5 type: chat max_input_tokens: 1000000 - input_price: 2.0 - output_price: 10.0 - cache_read_price: 0.2 - cache_write_price: 2.5 + input_price: 2.2 + output_price: 11.0 + cache_read_price: 0.22 + cache_write_price: 2.75 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3441,10 +3441,10 @@ - name: us.anthropic.claude-sonnet-5 type: chat max_input_tokens: 1000000 - input_price: 2.0 - output_price: 10.0 - cache_read_price: 0.2 - cache_write_price: 2.5 + input_price: 2.2 + output_price: 11.0 + cache_read_price: 0.22 + cache_write_price: 2.75 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -3507,10 +3507,10 @@ - name: us.anthropic.claude-fable-5 type: chat max_input_tokens: 1000000 - input_price: 10.0 - output_price: 50.0 - cache_read_price: 1.0 - cache_write_price: 12.5 + input_price: 11.0 + output_price: 55.0 + cache_read_price: 1.1 + cache_write_price: 13.75 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -3541,10 +3541,10 @@ - name: au.anthropic.claude-opus-4-8 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3589,10 +3589,10 @@ - name: jp.anthropic.claude-opus-4-8 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3605,10 +3605,10 @@ - name: us.anthropic.claude-opus-4-8 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -3716,10 +3716,10 @@ - name: jp.anthropic.claude-opus-4-7 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3732,10 +3732,10 @@ - name: us.anthropic.claude-opus-4-7 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -3805,11 +3805,11 @@ - name: anthropic.claude-sonnet-4-6 type: chat max_input_tokens: 1000000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 - max_output_tokens: 64000 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 + max_output_tokens: 128000 supports_vision: true supports_function_calling: true reasoning_levels: @@ -3839,7 +3839,7 @@ output_price: 16.5 cache_read_price: 0.33 cache_write_price: 4.125 - max_output_tokens: 64000 + max_output_tokens: 128000 supports_vision: true supports_function_calling: true reasoning_levels: @@ -3854,7 +3854,7 @@ output_price: 15.0 cache_read_price: 0.3 cache_write_price: 3.75 - max_output_tokens: 64000 + max_output_tokens: 128000 supports_vision: true supports_function_calling: true reasoning_levels: @@ -3865,11 +3865,11 @@ - name: jp.anthropic.claude-sonnet-4-6 type: chat max_input_tokens: 1000000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 - max_output_tokens: 64000 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 + max_output_tokens: 128000 supports_vision: true supports_function_calling: true reasoning_levels: @@ -3880,11 +3880,11 @@ - name: us.anthropic.claude-sonnet-4-6 type: chat max_input_tokens: 1000000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 - max_output_tokens: 64000 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 + max_output_tokens: 128000 require_max_tokens: true supports_vision: true supports_function_calling: true @@ -3898,10 +3898,10 @@ type: chat real_name: us.anthropic.claude-sonnet-4-6 max_input_tokens: 1000000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 patch: body: inferenceConfig: @@ -3932,10 +3932,10 @@ - name: anthropic.claude-opus-4-6-v1 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3947,10 +3947,10 @@ - name: au.anthropic.claude-opus-4-6-v1 type: chat max_input_tokens: 1000000 - input_price: 16.5 - output_price: 82.5 - cache_read_price: 1.65 - cache_write_price: 20.625 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 supports_vision: true supports_function_calling: true @@ -3992,10 +3992,10 @@ - name: us.anthropic.claude-opus-4-6-v1 type: chat max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 128000 require_max_tokens: true supports_vision: true @@ -4010,10 +4010,10 @@ type: chat real_name: us.anthropic.claude-opus-4-6-v1 max_input_tokens: 1000000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 patch: body: inferenceConfig: @@ -4186,6 +4186,13 @@ output_price: 1.85 max_output_tokens: 81920 supports_function_calling: true + - name: moonshot.kimi-k2-thinking + type: chat + max_input_tokens: 262143 + input_price: 0.6 + output_price: 2.5 + max_output_tokens: 16000 + supports_function_calling: true - name: anthropic.claude-opus-4-5-20251101-v1:0 type: chat max_input_tokens: 200000 @@ -4231,10 +4238,10 @@ - name: us.anthropic.claude-opus-4-5-20251101-v1:0 type: chat max_input_tokens: 200000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 max_output_tokens: 64000 require_max_tokens: true supports_vision: true @@ -4247,10 +4254,10 @@ type: chat real_name: us.anthropic.claude-opus-4-5-20251101-v1:0 max_input_tokens: 200000 - input_price: 5.0 - output_price: 25.0 - cache_read_price: 0.5 - cache_write_price: 6.25 + input_price: 5.5 + output_price: 27.5 + cache_read_price: 0.55 + cache_write_price: 6.875 patch: body: inferenceConfig: @@ -4263,13 +4270,6 @@ max_output_tokens: 24000 require_max_tokens: true supports_vision: true - - name: moonshot.kimi-k2-thinking - type: chat - max_input_tokens: 262143 - input_price: 0.6 - output_price: 2.5 - max_output_tokens: 16000 - supports_function_calling: true - name: openai.gpt-oss-safeguard-120b type: chat max_input_tokens: 128000 @@ -4320,10 +4320,10 @@ - name: au.anthropic.claude-haiku-4-5-20251001-v1:0 type: chat max_input_tokens: 200000 - input_price: 1.0 - output_price: 5.0 - cache_read_price: 0.1 - cache_write_price: 1.25 + input_price: 1.1 + output_price: 5.5 + cache_read_price: 0.11 + cache_write_price: 1.375 max_output_tokens: 64000 supports_vision: true supports_function_calling: true @@ -4350,20 +4350,20 @@ - name: jp.anthropic.claude-haiku-4-5-20251001-v1:0 type: chat max_input_tokens: 200000 - input_price: 1.0 - output_price: 5.0 - cache_read_price: 0.1 - cache_write_price: 1.25 + input_price: 1.1 + output_price: 5.5 + cache_read_price: 0.11 + cache_write_price: 1.375 max_output_tokens: 64000 supports_vision: true supports_function_calling: true - name: us.anthropic.claude-haiku-4-5-20251001-v1:0 type: chat max_input_tokens: 200000 - input_price: 1.0 - output_price: 5.0 - cache_read_price: 0.1 - cache_write_price: 1.25 + input_price: 1.1 + output_price: 5.5 + cache_read_price: 0.11 + cache_write_price: 1.375 max_output_tokens: 64000 require_max_tokens: true supports_vision: true @@ -4372,10 +4372,10 @@ type: chat real_name: us.anthropic.claude-haiku-4-5-20251001-v1:0 max_input_tokens: 200000 - input_price: 1.0 - output_price: 5.0 - cache_read_price: 0.1 - cache_write_price: 1.25 + input_price: 1.1 + output_price: 5.5 + cache_read_price: 0.11 + cache_write_price: 1.375 patch: body: inferenceConfig: @@ -4401,10 +4401,10 @@ - name: au.anthropic.claude-sonnet-4-5-20250929-v1:0 type: chat max_input_tokens: 200000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 max_output_tokens: 64000 supports_vision: true supports_function_calling: true @@ -4431,20 +4431,20 @@ - name: jp.anthropic.claude-sonnet-4-5-20250929-v1:0 type: chat max_input_tokens: 200000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 max_output_tokens: 64000 supports_vision: true supports_function_calling: true - name: us.anthropic.claude-sonnet-4-5-20250929-v1:0 type: chat max_input_tokens: 200000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 max_output_tokens: 64000 require_max_tokens: true supports_vision: true @@ -4453,10 +4453,10 @@ type: chat real_name: us.anthropic.claude-sonnet-4-5-20250929-v1:0 max_input_tokens: 200000 - input_price: 3.0 - output_price: 15.0 - cache_read_price: 0.3 - cache_write_price: 3.75 + input_price: 3.3 + output_price: 16.5 + cache_read_price: 0.33 + cache_write_price: 4.125 patch: body: inferenceConfig: @@ -4628,6 +4628,47 @@ output_price: 0.3 max_output_tokens: 8192 supports_function_calling: true + - name: apac.anthropic.claude-sonnet-4-20250514-v1:0 + type: chat + max_input_tokens: 200000 + input_price: 3.0 + output_price: 15.0 + cache_read_price: 0.3 + cache_write_price: 3.75 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + - name: eu.anthropic.claude-sonnet-4-20250514-v1:0 + type: chat + max_input_tokens: 200000 + input_price: 3.0 + output_price: 15.0 + cache_read_price: 0.3 + cache_write_price: 3.75 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + - name: global.anthropic.claude-sonnet-4-20250514-v1:0 + type: chat + max_input_tokens: 200000 + input_price: 3.0 + output_price: 15.0 + cache_read_price: 0.3 + cache_write_price: 3.75 + max_output_tokens: 64000 + supports_vision: true + supports_function_calling: true + - name: us.anthropic.claude-sonnet-4-20250514-v1:0 + type: chat + max_input_tokens: 200000 + input_price: 3.0 + output_price: 15.0 + cache_read_price: 0.3 + cache_write_price: 3.75 + max_output_tokens: 64000 + require_max_tokens: true + supports_vision: true + supports_function_calling: true - name: us.amazon.nova-premier-v1:0 type: chat max_input_tokens: 300000 @@ -6914,9 +6955,9 @@ - name: deepseek/deepseek-v4-flash-0731 type: chat max_input_tokens: 1310720 - input_price: 0.06 - output_price: 0.12 - cache_read_price: 0.012 + input_price: 0.055 + output_price: 0.11 + cache_read_price: 0.011 max_output_tokens: 943718 supports_function_calling: true - name: deepseek/deepseek-v4-flash-0731:batch @@ -7364,14 +7405,6 @@ max_output_tokens: 16384 supports_vision: true supports_function_calling: true - - name: google/gemma-4-31b-it:batch - type: chat - max_input_tokens: 262144 - input_price: 0.39 - output_price: 0.97 - max_output_tokens: 235929 - supports_vision: true - supports_function_calling: true - name: google/gemma-4-31b-it:free type: chat max_input_tokens: 262144 @@ -7670,7 +7703,7 @@ - name: minimax/minimax-m1 type: chat max_input_tokens: 1000000 - input_price: 0.55 + input_price: 0.4 output_price: 2.2 max_output_tokens: 40000 supports_function_calling: true @@ -7916,8 +7949,8 @@ - name: mistralai/mistral-small-3.2-24b-instruct type: chat max_input_tokens: 256000 - input_price: 0.075 - output_price: 0.2 + input_price: 0.0938 + output_price: 0.25 max_output_tokens: 16384 supports_vision: true supports_function_calling: true @@ -7980,9 +8013,9 @@ - name: moonshotai/kimi-k2.7-code type: chat max_input_tokens: 262144 - input_price: 0.71 - output_price: 3.5 - cache_read_price: 0.15 + input_price: 0.7062 + output_price: 3.21 + cache_read_price: 0.18 max_output_tokens: 235929 supports_vision: true supports_function_calling: true @@ -8083,10 +8116,10 @@ - name: nvidia/nemotron-3-ultra-550b-a55b type: chat max_input_tokens: 262144 - input_price: 0.6 - output_price: 2.4 - cache_read_price: 0.12 - max_output_tokens: 182520 + input_price: 0.625 + output_price: 3.125 + cache_read_price: 0.1875 + max_output_tokens: 32768 supports_function_calling: true - name: nvidia/nemotron-3-ultra-550b-a55b:free type: chat @@ -8969,12 +9002,6 @@ cache_read_price: 0.03 max_output_tokens: 117964 supports_function_calling: true - - name: openai/gpt-oss-20b:batch - type: chat - max_input_tokens: 131072 - input_price: 0.05 - output_price: 0.2 - max_output_tokens: 117964 - name: openai/gpt-oss-safeguard-20b type: chat max_input_tokens: 131072 @@ -9792,15 +9819,6 @@ max_output_tokens: 262144 supports_vision: true supports_function_calling: true - - name: thinkingmachines/inkling-small:batch - type: chat - max_input_tokens: 524288 - input_price: 0.5 - output_price: 1.2 - cache_read_price: 0.1 - max_output_tokens: 471859 - supports_vision: true - supports_function_calling: true - name: thinkingmachines/inkling-small:free type: chat max_input_tokens: 1048576 @@ -10030,6 +10048,12 @@ cache_read_price: 0.07 max_output_tokens: 943718 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 @@ -10041,9 +10065,9 @@ - name: z-ai/glm-5.3-flash type: chat max_input_tokens: 1310720 - input_price: 0.15 - output_price: 0.5 - cache_read_price: 0.03 + input_price: 0.1 + output_price: 0.3333 + cache_read_price: 0.02 max_output_tokens: 131072 supports_vision: true supports_function_calling: true @@ -10237,10 +10261,10 @@ - name: ~z-ai/glm-latest type: chat max_input_tokens: 1310720 - input_price: 0.8775 - output_price: 2.97 - cache_read_price: 0.1755 - max_output_tokens: 235929 + input_price: 0.85 + output_price: 2.8985 + cache_read_price: 0.17 + max_output_tokens: 943718 supports_function_calling: true - provider: github models: