chore(aihubmix): sync model catalog

providers/aihubmix/models/deepseek-v4-flash-think.toml → providers/aihubmix/models/alicloud-deepseek-v4-flash.toml

@@ -1,5 +1,5 @@
-name = "DeepSeek V4 Flash Think"
-family = "deepseek"
+name = "DeepSeek V4 Flash (Alibaba Cloud)"
+family = "deepseek-flash"
 release_date = "2026-04-24"
 last_updated = "2026-04-24"
 attachment = false
@@ -14,9 +14,9 @@ open_weights = true
 field = "reasoning_content"
 
 [cost]
-input = 0.154
-output = 0.308
-cache_read = 0.0308
+input = 0.14
+output = 0.28
+cache_read = 0.028
 
 [limit]
 context = 1_000_000

providers/aihubmix/models/alicloud-deepseek-v4-pro.toml

@@ -0,0 +1,27 @@
+name = "DeepSeek V4 Pro (Alibaba Cloud)"
+family = "deepseek-thinking"
+release_date = "2026-04-24"
+last_updated = "2026-04-24"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+knowledge = "2025-05"
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 1.69
+output = 3.38
+cache_read = 0.13
+
+[limit]
+context = 1_000_000
+output = 384_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/aihubmix/models/alicloud-glm-5.1.toml

@@ -0,0 +1,27 @@
+name = "GLM-5.1 (Alibaba Cloud)"
+family = "glm"
+release_date = "2026-03-27"
+last_updated = "2026-03-27"
+attachment = false
+reasoning = true
+temperature = true
+tool_call = true
+structured_output = true
+open_weights = true
+
+[interleaved]
+field = "reasoning_content"
+
+[cost]
+input = 0.84
+output = 3.38
+cache_read = 0.169
+cache_write = 1.05625
+
+[limit]
+context = 200_000
+output = 128_000
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/aihubmix/models/claude-opus-4-6-think.toml

@@ -1,23 +1,34 @@
-name = "Claude Opus 4.6"
+name = "Claude Opus 4.6 Thinking"
 family = "claude-opus"
 release_date = "2026-02-05"
 last_updated = "2026-03-13"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-05-31"
 open_weights = false
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 5
 output = 25
 cache_read = 0.5
 cache_write = 6.25
 
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 10
+output = 37.5
+cache_read = 1.0
+cache_write = 12.5
+
 [limit]
-context = 200_000
-output = 32_000
+context = 1_000_000
+output = 128_000
 
 [modalities]
 input = ["text", "image", "pdf"]

providers/aihubmix/models/claude-opus-4-6.toml

@@ -6,15 +6,25 @@ attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-05-31"
 open_weights = false
 
+interleaved = true
+
 [cost]
 input = 5
 output = 25
 cache_read = 0.5
 cache_write = 6.25
 
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 10
+output = 37.5
+cache_read = 1.0
+cache_write = 12.5
+
 [limit]
 context = 1_000_000
 output = 128_000

providers/aihubmix/models/claude-opus-4-7-think.toml

@@ -10,16 +10,26 @@ structured_output = true
 knowledge = "2026-01-31"
 open_weights = false
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 5
 output = 25
 cache_read = 0.5
 cache_write = 6.25
 
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 10
+output = 37.5
+cache_read = 1.0
+cache_write = 12.5
+
 [limit]
-context = 200_000
-output = 32_000
+context = 1_000_000
+output = 128_000
 
 [modalities]
-input = ["text", "image"]
+input = ["text", "image", "pdf"]
 output = ["text"]

providers/aihubmix/models/claude-opus-4-7.toml

@@ -6,15 +6,25 @@ attachment = true
 reasoning = true
 temperature = false
 tool_call = true
+structured_output = true
 knowledge = "2026-01-31"
 open_weights = false
 
+interleaved = true
+
 [cost]
 input = 5
 output = 25
 cache_read = 0.5
 cache_write = 6.25
 
+[[cost.tiers]]
+tier = { size = 200_000 }
+input = 10
+output = 37.5
+cache_read = 1.0
+cache_write = 12.5
+
 [limit]
 context = 1_000_000
 output = 128_000

providers/aihubmix/models/claude-sonnet-4-6-think.toml

@@ -1,14 +1,18 @@
-name = "Claude Sonnet 4.6 Think"
+name = "Claude Sonnet 4.6 Thinking"
 family = "claude-sonnet"
 release_date = "2026-02-17"
-last_updated = "2026-02-17"
+last_updated = "2026-03-13"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-08-31"
 open_weights = false
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 3.00
 output = 15.00
@@ -23,7 +27,7 @@ cache_read = 0.60
 cache_write = 7.50
 
 [limit]
-context = 200_000
+context = 1_000_000
 output = 64_000
 
 [modalities]

providers/aihubmix/models/claude-sonnet-4-6.toml

@@ -1,14 +1,17 @@
 name = "Claude Sonnet 4.6"
 family = "claude-sonnet"
 release_date = "2026-02-17"
-last_updated = "2026-02-17"
+last_updated = "2026-03-13"
 attachment = true
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 knowledge = "2025-08-31"
 open_weights = false
 
+interleaved = true
+
 [cost]
 input = 3.00
 output = 15.00
@@ -23,7 +26,7 @@ cache_read = 0.60
 cache_write = 7.50
 
 [limit]
-context = 200_000
+context = 1_000_000
 output = 64_000
 
 [modalities]

providers/aihubmix/models/coding-glm-5.1-free.toml

@@ -1,24 +1,23 @@
 name = "Coding GLM 5.1 (free)"
-family = "glm"
+family = "glm-free"
 release_date = "2026-04-11"
 last_updated = "2026-04-11"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
 structured_output = true
-open_weights = false
+open_weights = true
 
 [interleaved]
 field = "reasoning_content"
 
 [cost]
 input = 0
 output = 0
-cache_read = 0
 
 [limit]
-context = 204_800
+context = 200_000
 output = 128_000
 
 [modalities]

providers/aihubmix/models/coding-glm-5.1.toml

@@ -1,23 +1,25 @@
-name = "Coding-GLM-5.1"
+name = "Coding GLM 5.1"
 family = "glm"
 release_date = "2026-04-11"
 last_updated = "2026-04-11"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
-open_weights = false
+structured_output = true
+open_weights = true
 
 [interleaved]
 field = "reasoning_content"
 
 [cost]
 input = 0.06
 output = 0.22
+cache_read = 0.013
 
 [limit]
-context = 200000
-output = 128000
+context = 200_000
+output = 128_000
 
 [modalities]
 input = ["text"]

providers/aihubmix/models/coding-minimax-m2.7-free.toml

@@ -1,20 +1,24 @@
-name = "Coding-MiniMax-M2.7-Free"
-family = "minimax"
+name = "Coding MiniMax M2.7 (Free)"
+family = "minimax-free"
 release_date = "2026-03-18"
 last_updated = "2026-03-18"
 attachment = false
 reasoning = true
 temperature = true
 tool_call = true
+structured_output = true
 open_weights = true
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 0
 output = 0
 
 [limit]
 context = 204_800
-output = 13_100
+output = 128_100
 
 [modalities]
 input = ["text"]

providers/aihubmix/models/coding-minimax-m2.7-highspeed.toml

@@ -9,13 +9,16 @@ tool_call = true
 structured_output = true
 open_weights = true
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 0.2
 output = 0.2
 
 [limit]
 context = 204_800
-output = 13_100
+output = 128_100
 
 [modalities]
 input = ["text"]

providers/aihubmix/models/coding-minimax-m2.7.toml

@@ -9,13 +9,16 @@ tool_call = true
 structured_output = true
 open_weights = true
 
+[interleaved]
+field = "reasoning_content"
+
 [cost]
 input = 0.2
 output = 0.2
 
 [limit]
 context = 204_800
-output = 13_100
+output = 128_100
 
 [modalities]
 input = ["text"]

providers/aihubmix/models/coding-xiaomi-mimo-v2.5-pro.toml

@@ -0,0 +1,17 @@
+name = "Coding Xiaomi MiMo-V2.5-Pro"
+family = "mimo-v2.5-pro"
+last_updated = "2026-05-13"
+
+[extends]
+from = "xiaomi/mimo-v2.5-pro"
+
+[cost]
+input = 0.20
+output = 0.60
+cache_read = 0.04
+
+[[cost.tiers]]
+tier = { size = 256_000 }
+input = 0.40
+output = 1.20
+cache_read = 0.08