You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 

732 lines
26 KiB

  1. package ratio_setting
  2. import (
  3. "strings"
  4. "github.com/QuantumNous/new-api/common"
  5. "github.com/QuantumNous/new-api/setting/operation_setting"
  6. "github.com/QuantumNous/new-api/types"
  7. )
  8. // from songquanpeng/one-api
  9. const (
  10. USD2RMB = 7.3 // 暂定 1 USD = 7.3 RMB
  11. USD = 500 // $0.002 = 1 -> $1 = 500
  12. RMB = USD / USD2RMB
  13. )
  14. // modelRatio
  15. // https://platform.openai.com/docs/models/model-endpoint-compatibility
  16. // https://cloud.baidu.com/doc/WENXINWORKSHOP/s/Blfmc9dlf
  17. // https://openai.com/pricing
  18. // TODO: when a new api is enabled, check the pricing here
  19. // 1 === $0.002 / 1K tokens
  20. // 1 === ¥0.014 / 1k tokens
  21. var defaultModelRatio = map[string]float64{
  22. //"midjourney": 50,
  23. "gpt-4-gizmo-*": 15,
  24. "gpt-4o-gizmo-*": 2.5,
  25. "gpt-4-all": 15,
  26. "gpt-4o-all": 15,
  27. "gpt-4": 15,
  28. //"gpt-4-0314": 15, //deprecated
  29. "gpt-4-0613": 15,
  30. "gpt-4-32k": 30,
  31. //"gpt-4-32k-0314": 30, //deprecated
  32. "gpt-4-32k-0613": 30,
  33. "gpt-4-1106-preview": 5, // $10 / 1M tokens
  34. "gpt-4-0125-preview": 5, // $10 / 1M tokens
  35. "gpt-4-turbo-preview": 5, // $10 / 1M tokens
  36. "gpt-4-vision-preview": 5, // $10 / 1M tokens
  37. "gpt-4-1106-vision-preview": 5, // $10 / 1M tokens
  38. "chatgpt-4o-latest": 2.5, // $5 / 1M tokens
  39. "gpt-4o": 1.25, // $2.5 / 1M tokens
  40. "gpt-4o-audio-preview": 1.25, // $2.5 / 1M tokens
  41. "gpt-4o-audio-preview-2024-10-01": 1.25, // $2.5 / 1M tokens
  42. "gpt-4o-2024-05-13": 2.5, // $5 / 1M tokens
  43. "gpt-4o-2024-08-06": 1.25, // $2.5 / 1M tokens
  44. "gpt-4o-2024-11-20": 1.25, // $2.5 / 1M tokens
  45. "gpt-4o-realtime-preview": 2.5,
  46. "gpt-4o-realtime-preview-2024-10-01": 2.5,
  47. "gpt-4o-realtime-preview-2024-12-17": 2.5,
  48. "gpt-4o-mini-realtime-preview": 0.3,
  49. "gpt-4o-mini-realtime-preview-2024-12-17": 0.3,
  50. "gpt-4.1": 1.0, // $2 / 1M tokens
  51. "gpt-4.1-2025-04-14": 1.0, // $2 / 1M tokens
  52. "gpt-4.1-mini": 0.2, // $0.4 / 1M tokens
  53. "gpt-4.1-mini-2025-04-14": 0.2, // $0.4 / 1M tokens
  54. "gpt-4.1-nano": 0.05, // $0.1 / 1M tokens
  55. "gpt-4.1-nano-2025-04-14": 0.05, // $0.1 / 1M tokens
  56. "gpt-image-1": 2.5, // $5 / 1M tokens
  57. "o1": 7.5, // $15 / 1M tokens
  58. "o1-2024-12-17": 7.5, // $15 / 1M tokens
  59. "o1-preview": 7.5, // $15 / 1M tokens
  60. "o1-preview-2024-09-12": 7.5, // $15 / 1M tokens
  61. "o1-mini": 0.55, // $1.1 / 1M tokens
  62. "o1-mini-2024-09-12": 0.55, // $1.1 / 1M tokens
  63. "o1-pro": 75.0, // $150 / 1M tokens
  64. "o1-pro-2025-03-19": 75.0, // $150 / 1M tokens
  65. "o3-mini": 0.55,
  66. "o3-mini-2025-01-31": 0.55,
  67. "o3-mini-high": 0.55,
  68. "o3-mini-2025-01-31-high": 0.55,
  69. "o3-mini-low": 0.55,
  70. "o3-mini-2025-01-31-low": 0.55,
  71. "o3-mini-medium": 0.55,
  72. "o3-mini-2025-01-31-medium": 0.55,
  73. "o3": 1.0, // $2 / 1M tokens
  74. "o3-2025-04-16": 1.0, // $2 / 1M tokens
  75. "o3-pro": 10.0, // $20 / 1M tokens
  76. "o3-pro-2025-06-10": 10.0, // $20 / 1M tokens
  77. "o3-deep-research": 5.0, // $10 / 1M tokens
  78. "o3-deep-research-2025-06-26": 5.0, // $10 / 1M tokens
  79. "o4-mini": 0.55, // $1.1 / 1M tokens
  80. "o4-mini-2025-04-16": 0.55, // $1.1 / 1M tokens
  81. "o4-mini-deep-research": 1.0, // $2 / 1M tokens
  82. "o4-mini-deep-research-2025-06-26": 1.0, // $2 / 1M tokens
  83. "gpt-4o-mini": 0.075,
  84. "gpt-4o-mini-2024-07-18": 0.075,
  85. "gpt-4-turbo": 5, // $0.01 / 1K tokens
  86. "gpt-4-turbo-2024-04-09": 5, // $0.01 / 1K tokens
  87. "gpt-4.5-preview": 37.5,
  88. "gpt-4.5-preview-2025-02-27": 37.5,
  89. "gpt-5": 0.625,
  90. "gpt-5-2025-08-07": 0.625,
  91. "gpt-5-chat-latest": 0.625,
  92. "gpt-5-mini": 0.125,
  93. "gpt-5-mini-2025-08-07": 0.125,
  94. "gpt-5-nano": 0.025,
  95. "gpt-5-nano-2025-08-07": 0.025,
  96. //"gpt-3.5-turbo-0301": 0.75, //deprecated
  97. "gpt-3.5-turbo": 0.25,
  98. "gpt-3.5-turbo-0613": 0.75,
  99. "gpt-3.5-turbo-16k": 1.5, // $0.003 / 1K tokens
  100. "gpt-3.5-turbo-16k-0613": 1.5,
  101. "gpt-3.5-turbo-instruct": 0.75, // $0.0015 / 1K tokens
  102. "gpt-3.5-turbo-1106": 0.5, // $0.001 / 1K tokens
  103. "gpt-3.5-turbo-0125": 0.25,
  104. "babbage-002": 0.2, // $0.0004 / 1K tokens
  105. "davinci-002": 1, // $0.002 / 1K tokens
  106. "text-ada-001": 0.2,
  107. "text-babbage-001": 0.25,
  108. "text-curie-001": 1,
  109. //"text-davinci-002": 10,
  110. //"text-davinci-003": 10,
  111. "text-davinci-edit-001": 10,
  112. "code-davinci-edit-001": 10,
  113. "whisper-1": 15, // $0.006 / minute -> $0.006 / 150 words -> $0.006 / 200 tokens -> $0.03 / 1k tokens
  114. "tts-1": 7.5, // 1k characters -> $0.015
  115. "tts-1-1106": 7.5, // 1k characters -> $0.015
  116. "tts-1-hd": 15, // 1k characters -> $0.03
  117. "tts-1-hd-1106": 15, // 1k characters -> $0.03
  118. "davinci": 10,
  119. "curie": 10,
  120. "babbage": 10,
  121. "ada": 10,
  122. "text-embedding-3-small": 0.01,
  123. "text-embedding-3-large": 0.065,
  124. "text-embedding-ada-002": 0.05,
  125. "text-search-ada-doc-001": 10,
  126. "text-moderation-stable": 0.1,
  127. "text-moderation-latest": 0.1,
  128. "claude-3-haiku-20240307": 0.125, // $0.25 / 1M tokens
  129. "claude-3-5-haiku-20241022": 0.5, // $1 / 1M tokens
  130. "claude-haiku-4-5-20251001": 0.5, // $1 / 1M tokens
  131. "claude-3-sonnet-20240229": 1.5, // $3 / 1M tokens
  132. "claude-3-5-sonnet-20240620": 1.5,
  133. "claude-3-5-sonnet-20241022": 1.5,
  134. "claude-3-7-sonnet-20250219": 1.5,
  135. "claude-3-7-sonnet-20250219-thinking": 1.5,
  136. "claude-sonnet-4-20250514": 1.5,
  137. "claude-sonnet-4-5-20250929": 1.5,
  138. "claude-opus-4-5-20251101": 2.5,
  139. "claude-opus-4-6": 2.5,
  140. "claude-opus-4-6-max": 2.5,
  141. "claude-opus-4-6-high": 2.5,
  142. "claude-opus-4-6-medium": 2.5,
  143. "claude-opus-4-6-low": 2.5,
  144. "claude-3-opus-20240229": 7.5, // $15 / 1M tokens
  145. "claude-opus-4-20250514": 7.5,
  146. "claude-opus-4-1-20250805": 7.5,
  147. "ERNIE-4.0-8K": 0.120 * RMB,
  148. "ERNIE-3.5-8K": 0.012 * RMB,
  149. "ERNIE-3.5-8K-0205": 0.024 * RMB,
  150. "ERNIE-3.5-8K-1222": 0.012 * RMB,
  151. "ERNIE-Bot-8K": 0.024 * RMB,
  152. "ERNIE-3.5-4K-0205": 0.012 * RMB,
  153. "ERNIE-Speed-8K": 0.004 * RMB,
  154. "ERNIE-Speed-128K": 0.004 * RMB,
  155. "ERNIE-Lite-8K-0922": 0.008 * RMB,
  156. "ERNIE-Lite-8K-0308": 0.003 * RMB,
  157. "ERNIE-Tiny-8K": 0.001 * RMB,
  158. "BLOOMZ-7B": 0.004 * RMB,
  159. "Embedding-V1": 0.002 * RMB,
  160. "bge-large-zh": 0.002 * RMB,
  161. "bge-large-en": 0.002 * RMB,
  162. "tao-8k": 0.002 * RMB,
  163. "PaLM-2": 1,
  164. "gemini-1.5-pro-latest": 1.25, // $3.5 / 1M tokens
  165. "gemini-1.5-flash-latest": 0.075,
  166. "gemini-2.0-flash": 0.05,
  167. "gemini-2.5-pro-exp-03-25": 0.625,
  168. "gemini-2.5-pro-preview-03-25": 0.625,
  169. "gemini-2.5-pro": 0.625,
  170. "gemini-2.5-flash-preview-04-17": 0.075,
  171. "gemini-2.5-flash-preview-04-17-thinking": 0.075,
  172. "gemini-2.5-flash-preview-04-17-nothinking": 0.075,
  173. "gemini-2.5-flash-preview-05-20": 0.075,
  174. "gemini-2.5-flash-preview-05-20-thinking": 0.075,
  175. "gemini-2.5-flash-preview-05-20-nothinking": 0.075,
  176. "gemini-2.5-flash-thinking-*": 0.075, // 用于为后续所有2.5 flash thinking budget 模型设置默认倍率
  177. "gemini-2.5-pro-thinking-*": 0.625, // 用于为后续所有2.5 pro thinking budget 模型设置默认倍率
  178. "gemini-2.5-flash-lite-preview-thinking-*": 0.05,
  179. "gemini-2.5-flash-lite-preview-06-17": 0.05,
  180. "gemini-2.5-flash": 0.15,
  181. "gemini-robotics-er-1.5-preview": 0.15,
  182. "gemini-embedding-001": 0.075,
  183. "text-embedding-004": 0.001,
  184. "chatglm_turbo": 0.3572, // ¥0.005 / 1k tokens
  185. "chatglm_pro": 0.7143, // ¥0.01 / 1k tokens
  186. "chatglm_std": 0.3572, // ¥0.005 / 1k tokens
  187. "chatglm_lite": 0.1429, // ¥0.002 / 1k tokens
  188. "glm-4": 7.143, // ¥0.1 / 1k tokens
  189. "glm-4v": 0.05 * RMB, // ¥0.05 / 1k tokens
  190. "glm-4-alltools": 0.1 * RMB, // ¥0.1 / 1k tokens
  191. "glm-3-turbo": 0.3572,
  192. "glm-4-plus": 0.05 * RMB,
  193. "glm-4-0520": 0.1 * RMB,
  194. "glm-4-air": 0.001 * RMB,
  195. "glm-4-airx": 0.01 * RMB,
  196. "glm-4-long": 0.001 * RMB,
  197. "glm-4-flash": 0,
  198. "glm-4v-plus": 0.01 * RMB,
  199. "qwen-turbo": 0.8572, // ¥0.012 / 1k tokens
  200. "qwen-plus": 10, // ¥0.14 / 1k tokens
  201. "text-embedding-v1": 0.05, // ¥0.0007 / 1k tokens
  202. "SparkDesk-v1.1": 1.2858, // ¥0.018 / 1k tokens
  203. "SparkDesk-v2.1": 1.2858, // ¥0.018 / 1k tokens
  204. "SparkDesk-v3.1": 1.2858, // ¥0.018 / 1k tokens
  205. "SparkDesk-v3.5": 1.2858, // ¥0.018 / 1k tokens
  206. "SparkDesk-v4.0": 1.2858,
  207. "360GPT_S2_V9": 0.8572, // ¥0.012 / 1k tokens
  208. "360gpt-turbo": 0.0858, // ¥0.0012 / 1k tokens
  209. "360gpt-turbo-responsibility-8k": 0.8572, // ¥0.012 / 1k tokens
  210. "360gpt-pro": 0.8572, // ¥0.012 / 1k tokens
  211. "360gpt2-pro": 0.8572, // ¥0.012 / 1k tokens
  212. "embedding-bert-512-v1": 0.0715, // ¥0.001 / 1k tokens
  213. "embedding_s1_v1": 0.0715, // ¥0.001 / 1k tokens
  214. "semantic_similarity_s1_v1": 0.0715, // ¥0.001 / 1k tokens
  215. "hunyuan": 7.143, // ¥0.1 / 1k tokens // https://cloud.tencent.com/document/product/1729/97731#e0e6be58-60c8-469f-bdeb-6c264ce3b4d0
  216. // https://platform.lingyiwanwu.com/docs#-计费单元
  217. // 已经按照 7.2 来换算美元价格
  218. "yi-34b-chat-0205": 0.18,
  219. "yi-34b-chat-200k": 0.864,
  220. "yi-vl-plus": 0.432,
  221. "yi-large": 20.0 / 1000 * RMB,
  222. "yi-medium": 2.5 / 1000 * RMB,
  223. "yi-vision": 6.0 / 1000 * RMB,
  224. "yi-medium-200k": 12.0 / 1000 * RMB,
  225. "yi-spark": 1.0 / 1000 * RMB,
  226. "yi-large-rag": 25.0 / 1000 * RMB,
  227. "yi-large-turbo": 12.0 / 1000 * RMB,
  228. "yi-large-preview": 20.0 / 1000 * RMB,
  229. "yi-large-rag-preview": 25.0 / 1000 * RMB,
  230. "command": 0.5,
  231. "command-nightly": 0.5,
  232. "command-light": 0.5,
  233. "command-light-nightly": 0.5,
  234. "command-r": 0.25,
  235. "command-r-plus": 1.5,
  236. "command-r-08-2024": 0.075,
  237. "command-r-plus-08-2024": 1.25,
  238. "deepseek-chat": 0.27 / 2,
  239. "deepseek-coder": 0.27 / 2,
  240. "deepseek-reasoner": 0.55 / 2, // 0.55 / 1k tokens
  241. // Perplexity online 模型对搜索额外收费,有需要应自行调整,此处不计入搜索费用
  242. "llama-3-sonar-small-32k-chat": 0.2 / 1000 * USD,
  243. "llama-3-sonar-small-32k-online": 0.2 / 1000 * USD,
  244. "llama-3-sonar-large-32k-chat": 1 / 1000 * USD,
  245. "llama-3-sonar-large-32k-online": 1 / 1000 * USD,
  246. // grok
  247. "grok-3-beta": 1.5,
  248. "grok-3-mini-beta": 0.15,
  249. "grok-2": 1,
  250. "grok-2-vision": 1,
  251. "grok-beta": 2.5,
  252. "grok-vision-beta": 2.5,
  253. "grok-3-fast-beta": 2.5,
  254. "grok-3-mini-fast-beta": 0.3,
  255. // submodel
  256. "NousResearch/Hermes-4-405B-FP8": 0.8,
  257. "Qwen/Qwen3-235B-A22B-Thinking-2507": 0.6,
  258. "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8": 0.8,
  259. "Qwen/Qwen3-235B-A22B-Instruct-2507": 0.3,
  260. "zai-org/GLM-4.5-FP8": 0.8,
  261. "openai/gpt-oss-120b": 0.5,
  262. "deepseek-ai/DeepSeek-R1-0528": 0.8,
  263. "deepseek-ai/DeepSeek-R1": 0.8,
  264. "deepseek-ai/DeepSeek-V3-0324": 0.8,
  265. "deepseek-ai/DeepSeek-V3.1": 0.8,
  266. }
  267. var defaultModelPrice = map[string]float64{
  268. "suno_music": 0.1,
  269. "suno_lyrics": 0.01,
  270. "dall-e-3": 0.04,
  271. "imagen-3.0-generate-002": 0.03,
  272. "black-forest-labs/flux-1.1-pro": 0.04,
  273. "gpt-4-gizmo-*": 0.1,
  274. "mj_video": 0.8,
  275. "mj_imagine": 0.1,
  276. "mj_edits": 0.1,
  277. "mj_variation": 0.1,
  278. "mj_reroll": 0.1,
  279. "mj_blend": 0.1,
  280. "mj_modal": 0.1,
  281. "mj_zoom": 0.1,
  282. "mj_shorten": 0.1,
  283. "mj_high_variation": 0.1,
  284. "mj_low_variation": 0.1,
  285. "mj_pan": 0.1,
  286. "mj_inpaint": 0,
  287. "mj_custom_zoom": 0,
  288. "mj_describe": 0.05,
  289. "mj_upscale": 0.05,
  290. "swap_face": 0.05,
  291. "mj_upload": 0.05,
  292. "sora-2": 0.3,
  293. "sora-2-pro": 0.5,
  294. "gpt-4o-mini-tts": 0.3,
  295. }
  296. var defaultAudioRatio = map[string]float64{
  297. "gpt-4o-audio-preview": 16,
  298. "gpt-4o-mini-audio-preview": 66.67,
  299. "gpt-4o-realtime-preview": 8,
  300. "gpt-4o-mini-realtime-preview": 16.67,
  301. "gpt-4o-mini-tts": 25,
  302. }
  303. var defaultAudioCompletionRatio = map[string]float64{
  304. "gpt-4o-realtime": 2,
  305. "gpt-4o-mini-realtime": 2,
  306. "gpt-4o-mini-tts": 1,
  307. "tts-1": 0,
  308. "tts-1-hd": 0,
  309. "tts-1-1106": 0,
  310. "tts-1-hd-1106": 0,
  311. }
  312. var modelPriceMap = types.NewRWMap[string, float64]()
  313. var modelRatioMap = types.NewRWMap[string, float64]()
  314. var completionRatioMap = types.NewRWMap[string, float64]()
  315. var defaultCompletionRatio = map[string]float64{
  316. "gpt-4-gizmo-*": 2,
  317. "gpt-4o-gizmo-*": 3,
  318. "gpt-4-all": 2,
  319. "gpt-image-1": 8,
  320. }
  321. // InitRatioSettings initializes all model related settings maps
  322. func InitRatioSettings() {
  323. modelPriceMap.AddAll(defaultModelPrice)
  324. modelRatioMap.AddAll(defaultModelRatio)
  325. completionRatioMap.AddAll(defaultCompletionRatio)
  326. cacheRatioMap.AddAll(defaultCacheRatio)
  327. createCacheRatioMap.AddAll(defaultCreateCacheRatio)
  328. imageRatioMap.AddAll(defaultImageRatio)
  329. audioRatioMap.AddAll(defaultAudioRatio)
  330. audioCompletionRatioMap.AddAll(defaultAudioCompletionRatio)
  331. }
  332. func GetModelPriceMap() map[string]float64 {
  333. return modelPriceMap.ReadAll()
  334. }
  335. func ModelPrice2JSONString() string {
  336. return modelPriceMap.MarshalJSONString()
  337. }
  338. func UpdateModelPriceByJSONString(jsonStr string) error {
  339. return types.LoadFromJsonStringWithCallback(modelPriceMap, jsonStr, InvalidateExposedDataCache)
  340. }
  341. // GetModelPrice 返回模型的价格,如果模型不存在则返回-1,false
  342. func GetModelPrice(name string, printErr bool) (float64, bool) {
  343. name = FormatMatchingModelName(name)
  344. // 优先匹配精确模型名称
  345. if price, ok := modelPriceMap.Get(name); ok {
  346. return price, true
  347. }
  348. if strings.HasSuffix(name, CompactModelSuffix) {
  349. price, ok := modelPriceMap.Get(CompactWildcardModelKey)
  350. if !ok {
  351. if printErr {
  352. common.SysError("model price not found: " + name)
  353. }
  354. return -1, false
  355. }
  356. return price, true
  357. }
  358. price, ok := modelPriceMap.Get(name)
  359. if !ok {
  360. // 新增:如果模型名称包含 "/",尝试去掉供应商前缀后重新查找
  361. // 例如: "openai/gpt-4o" -> "gpt-4o"
  362. if strings.Contains(name, "/") {
  363. parts := strings.SplitN(name, "/", 2)
  364. if len(parts) == 2 {
  365. baseName := FormatMatchingModelName(parts[1])
  366. if basePrice, ok := modelPriceMap.Get(baseName); ok {
  367. return basePrice, true
  368. }
  369. }
  370. }
  371. if printErr {
  372. common.SysError("model price not found: " + name)
  373. }
  374. return -1, false
  375. }
  376. return price, true
  377. }
  378. func UpdateModelRatioByJSONString(jsonStr string) error {
  379. return types.LoadFromJsonStringWithCallback(modelRatioMap, jsonStr, InvalidateExposedDataCache)
  380. }
  381. // 处理带有思考预算的模型名称,方便统一定价
  382. func handleThinkingBudgetModel(name, prefix, wildcard string) string {
  383. if strings.HasPrefix(name, prefix) && strings.Contains(name, "-thinking-") {
  384. return wildcard
  385. }
  386. return name
  387. }
  388. func GetModelRatio(name string) (float64, bool, string) {
  389. name = FormatMatchingModelName(name)
  390. ratio, ok := modelRatioMap.Get(name)
  391. if !ok {
  392. if strings.HasSuffix(name, CompactModelSuffix) {
  393. if wildcardRatio, ok := modelRatioMap.Get(CompactWildcardModelKey); ok {
  394. return wildcardRatio, true, name
  395. }
  396. //return 0, true, name
  397. }
  398. // 新增:如果模型名称包含 "/",尝试去掉供应商前缀后重新查找
  399. // 例如: "openai/o3-mini" -> "o3-mini"
  400. if strings.Contains(name, "/") {
  401. parts := strings.SplitN(name, "/", 2)
  402. if len(parts) == 2 {
  403. baseName := FormatMatchingModelName(parts[1])
  404. if baseRatio, ok := modelRatioMap.Get(baseName); ok {
  405. return baseRatio, true, name
  406. }
  407. }
  408. }
  409. return 37.5, operation_setting.SelfUseModeEnabled, name
  410. }
  411. return ratio, true, name
  412. }
  413. func DefaultModelRatio2JSONString() string {
  414. jsonBytes, err := common.Marshal(defaultModelRatio)
  415. if err != nil {
  416. common.SysError("error marshalling model ratio: " + err.Error())
  417. }
  418. return string(jsonBytes)
  419. }
  420. func GetDefaultModelRatioMap() map[string]float64 {
  421. return defaultModelRatio
  422. }
  423. func GetDefaultModelPriceMap() map[string]float64 {
  424. return defaultModelPrice
  425. }
  426. func CompletionRatio2JSONString() string {
  427. return completionRatioMap.MarshalJSONString()
  428. }
  429. func UpdateCompletionRatioByJSONString(jsonStr string) error {
  430. return types.LoadFromJsonStringWithCallback(completionRatioMap, jsonStr, InvalidateExposedDataCache)
  431. }
  432. func GetCompletionRatio(name string) float64 {
  433. name = FormatMatchingModelName(name)
  434. if strings.Contains(name, "/") {
  435. if ratio, ok := completionRatioMap.Get(name); ok {
  436. return ratio
  437. }
  438. }
  439. hardCodedRatio, contain := getHardcodedCompletionModelRatio(name)
  440. if contain {
  441. return hardCodedRatio
  442. }
  443. if ratio, ok := completionRatioMap.Get(name); ok {
  444. return ratio
  445. }
  446. return hardCodedRatio
  447. }
  448. func getHardcodedCompletionModelRatio(name string) (float64, bool) {
  449. isReservedModel := strings.HasSuffix(name, "-all") || strings.HasSuffix(name, "-gizmo-*")
  450. if isReservedModel {
  451. return 2, false
  452. }
  453. if strings.HasPrefix(name, "gpt-") {
  454. if strings.HasPrefix(name, "gpt-4o") {
  455. if name == "gpt-4o-2024-05-13" {
  456. return 3, true
  457. }
  458. if strings.HasPrefix(name, "gpt-4o-mini-tts") {
  459. return 20, false
  460. }
  461. return 4, false
  462. }
  463. // gpt-5 匹配
  464. if strings.HasPrefix(name, "gpt-5") {
  465. return 8, true
  466. }
  467. // gpt-4.5-preview匹配
  468. if strings.HasPrefix(name, "gpt-4.5-preview") {
  469. return 2, true
  470. }
  471. if strings.HasPrefix(name, "gpt-4-turbo") || strings.HasSuffix(name, "gpt-4-1106") || strings.HasSuffix(name, "gpt-4-1105") {
  472. return 3, true
  473. }
  474. // 没有特殊标记的 gpt-4 模型默认倍率为 2
  475. return 2, false
  476. }
  477. if strings.HasPrefix(name, "o1") || strings.HasPrefix(name, "o3") {
  478. return 4, true
  479. }
  480. if name == "chatgpt-4o-latest" {
  481. return 3, true
  482. }
  483. if strings.Contains(name, "claude-3") {
  484. return 5, true
  485. } else if strings.Contains(name, "claude-sonnet-4") || strings.Contains(name, "claude-opus-4") || strings.Contains(name, "claude-haiku-4") {
  486. return 5, true
  487. }
  488. if strings.HasPrefix(name, "gpt-3.5") {
  489. if name == "gpt-3.5-turbo" || strings.HasSuffix(name, "0125") {
  490. // https://openai.com/blog/new-embedding-models-and-api-updates
  491. // Updated GPT-3.5 Turbo model and lower pricing
  492. return 3, true
  493. }
  494. if strings.HasSuffix(name, "1106") {
  495. return 2, true
  496. }
  497. return 4.0 / 3.0, true
  498. }
  499. if strings.HasPrefix(name, "mistral-") {
  500. return 3, true
  501. }
  502. if strings.HasPrefix(name, "gemini-") {
  503. if strings.HasPrefix(name, "gemini-1.5") {
  504. return 4, true
  505. } else if strings.HasPrefix(name, "gemini-2.0") {
  506. return 4, true
  507. } else if strings.HasPrefix(name, "gemini-2.5-pro") { // 移除preview来增加兼容性,这里假设正式版的倍率和preview一致
  508. return 8, false
  509. } else if strings.HasPrefix(name, "gemini-2.5-flash") { // 处理不同的flash模型倍率
  510. if strings.HasPrefix(name, "gemini-2.5-flash-preview") {
  511. if strings.HasSuffix(name, "-nothinking") {
  512. return 4, false
  513. }
  514. return 3.5 / 0.15, false
  515. }
  516. if strings.HasPrefix(name, "gemini-2.5-flash-lite") {
  517. return 4, false
  518. }
  519. return 2.5 / 0.3, false
  520. } else if strings.HasPrefix(name, "gemini-robotics-er-1.5") {
  521. return 2.5 / 0.3, false
  522. } else if strings.HasPrefix(name, "gemini-3-pro") {
  523. if strings.HasPrefix(name, "gemini-3-pro-image") {
  524. return 60, false
  525. }
  526. return 6, false
  527. }
  528. return 4, false
  529. }
  530. if strings.HasPrefix(name, "command") {
  531. switch name {
  532. case "command-r":
  533. return 3, true
  534. case "command-r-plus":
  535. return 5, true
  536. case "command-r-08-2024":
  537. return 4, true
  538. case "command-r-plus-08-2024":
  539. return 4, true
  540. default:
  541. return 4, false
  542. }
  543. }
  544. // hint 只给官方上4倍率,由于开源模型供应商自行定价,不对其进行补全倍率进行强制对齐
  545. if strings.HasPrefix(name, "ERNIE-Speed-") {
  546. return 2, true
  547. } else if strings.HasPrefix(name, "ERNIE-Lite-") {
  548. return 2, true
  549. } else if strings.HasPrefix(name, "ERNIE-Character") {
  550. return 2, true
  551. } else if strings.HasPrefix(name, "ERNIE-Functions") {
  552. return 2, true
  553. }
  554. switch name {
  555. case "llama2-70b-4096":
  556. return 0.8 / 0.64, true
  557. case "llama3-8b-8192":
  558. return 2, true
  559. case "llama3-70b-8192":
  560. return 0.79 / 0.59, true
  561. }
  562. return 1, false
  563. }
  564. func GetAudioRatio(name string) float64 {
  565. name = FormatMatchingModelName(name)
  566. if ratio, ok := audioRatioMap.Get(name); ok {
  567. return ratio
  568. }
  569. return 1
  570. }
  571. func GetAudioRatioV2(name string) (float64, bool) {
  572. name = FormatMatchingModelName(name)
  573. ratio, ok := audioRatioMap.Get(name)
  574. if !ok {
  575. return 0, false
  576. }
  577. return ratio, true
  578. }
  579. func GetAudioCompletionRatio(name string) float64 {
  580. name = FormatMatchingModelName(name)
  581. if ratio, ok := audioCompletionRatioMap.Get(name); ok {
  582. return ratio
  583. }
  584. return 1
  585. }
  586. func GetAudioCompletionRatioV2(name string) (float64, bool) {
  587. name = FormatMatchingModelName(name)
  588. ratio, ok := audioCompletionRatioMap.Get(name)
  589. if !ok {
  590. return 0, false
  591. }
  592. return ratio, true
  593. }
  594. func ContainsAudioRatio(name string) bool {
  595. name = FormatMatchingModelName(name)
  596. _, ok := audioRatioMap.Get(name)
  597. return ok
  598. }
  599. func ContainsAudioCompletionRatio(name string) bool {
  600. name = FormatMatchingModelName(name)
  601. _, ok := audioCompletionRatioMap.Get(name)
  602. return ok
  603. }
  604. func ModelRatio2JSONString() string {
  605. return modelRatioMap.MarshalJSONString()
  606. }
  607. var defaultImageRatio = map[string]float64{
  608. "gpt-image-1": 2,
  609. }
  610. var imageRatioMap = types.NewRWMap[string, float64]()
  611. var audioRatioMap = types.NewRWMap[string, float64]()
  612. var audioCompletionRatioMap = types.NewRWMap[string, float64]()
  613. func ImageRatio2JSONString() string {
  614. return imageRatioMap.MarshalJSONString()
  615. }
  616. func UpdateImageRatioByJSONString(jsonStr string) error {
  617. return types.LoadFromJsonString(imageRatioMap, jsonStr)
  618. }
  619. func GetImageRatio(name string) (float64, bool) {
  620. ratio, ok := imageRatioMap.Get(name)
  621. if !ok {
  622. return 1, false // Default to 1 if not found
  623. }
  624. return ratio, true
  625. }
  626. func AudioRatio2JSONString() string {
  627. return audioRatioMap.MarshalJSONString()
  628. }
  629. func UpdateAudioRatioByJSONString(jsonStr string) error {
  630. return types.LoadFromJsonStringWithCallback(audioRatioMap, jsonStr, InvalidateExposedDataCache)
  631. }
  632. func AudioCompletionRatio2JSONString() string {
  633. return audioCompletionRatioMap.MarshalJSONString()
  634. }
  635. func UpdateAudioCompletionRatioByJSONString(jsonStr string) error {
  636. return types.LoadFromJsonStringWithCallback(audioCompletionRatioMap, jsonStr, InvalidateExposedDataCache)
  637. }
  638. func GetModelRatioCopy() map[string]float64 {
  639. return modelRatioMap.ReadAll()
  640. }
  641. func GetModelPriceCopy() map[string]float64 {
  642. return modelPriceMap.ReadAll()
  643. }
  644. func GetCompletionRatioCopy() map[string]float64 {
  645. return completionRatioMap.ReadAll()
  646. }
  647. // 转换模型名,减少渠道必须配置各种带参数模型
  648. func FormatMatchingModelName(name string) string {
  649. if strings.HasPrefix(name, "gemini-2.5-flash-lite") {
  650. name = handleThinkingBudgetModel(name, "gemini-2.5-flash-lite", "gemini-2.5-flash-lite-thinking-*")
  651. } else if strings.HasPrefix(name, "gemini-2.5-flash") {
  652. name = handleThinkingBudgetModel(name, "gemini-2.5-flash", "gemini-2.5-flash-thinking-*")
  653. } else if strings.HasPrefix(name, "gemini-2.5-pro") {
  654. name = handleThinkingBudgetModel(name, "gemini-2.5-pro", "gemini-2.5-pro-thinking-*")
  655. }
  656. if strings.HasPrefix(name, "gpt-4-gizmo") {
  657. name = "gpt-4-gizmo-*"
  658. }
  659. if strings.HasPrefix(name, "gpt-4o-gizmo") {
  660. name = "gpt-4o-gizmo-*"
  661. }
  662. return name
  663. }
  664. // result: 倍率or价格, usePrice, exist
  665. func GetModelRatioOrPrice(model string) (float64, bool, bool) { // price or ratio
  666. price, usePrice := GetModelPrice(model, false)
  667. if usePrice {
  668. return price, true, true
  669. }
  670. modelRatio, success, _ := GetModelRatio(model)
  671. if success {
  672. return modelRatio, false, true
  673. }
  674. return 37.5, false, false
  675. }