Skip to content

Commit 3714f72

Browse files
mini2smini2s
authored
feat(types): add deepseek v4, kimi k2.6, and update codex gpt-5.5 defaults (#1163)
1 parent f9282f5 commit 3714f72

4 files changed

Lines changed: 65 additions & 16 deletions

File tree

‎packages/types/src/providers/deepseek.ts‎

Lines changed: 24 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -32,6 +32,30 @@ export const deepSeekModels = {
3232
cacheReadsPrice: 0.028, // $0.028 per million tokens (cache hit) - Updated Dec 9, 2025
3333
description: `DeepSeek-V3.2 (Thinking Mode) achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks. Supports Chain of Thought reasoning with up to 8K output tokens. Supports JSON output, tool calls, and chat prefix completion (beta).`,
3434
},
35+
"deepseek-v4-pro": {
36+
maxTokens: 384_000, // 384K max output
37+
contextWindow: 1_000_000,
38+
supportsImages: false,
39+
supportsPromptCache: true,
40+
preserveReasoning: true,
41+
inputPrice: 1.74, // $1.74 per million tokens (cache miss)
42+
outputPrice: 3.48, // $3.48 per million tokens
43+
cacheWritesPrice: 1.74, // $1.74 per million tokens (cache miss)
44+
cacheReadsPrice: 0.0145, // $0.0145 per million tokens (cache hit)
45+
description: `DeepSeek-V4-Pro is the flagship Mixture-of-Experts model in the DeepSeek-V4 Preview series with 1.6T total parameters and 49B activated parameters. Supports a 1M-token context window with up to 384K output tokens, thinking mode (default), JSON output, tool calls, chat prefix completion (beta), and FIM completion (beta, non-thinking only). Best for complex reasoning, advanced coding, agentic workflows, and long-context analysis.`,
46+
},
47+
"deepseek-v4-flash": {
48+
maxTokens: 384_000, // 384K max output
49+
contextWindow: 1_000_000,
50+
supportsImages: false,
51+
supportsPromptCache: true,
52+
preserveReasoning: true,
53+
inputPrice: 0.14, // $0.14 per million tokens (cache miss)
54+
outputPrice: 0.28, // $0.28 per million tokens
55+
cacheWritesPrice: 0.14, // $0.14 per million tokens (cache miss)
56+
cacheReadsPrice: 0.0028, // $0.0028 per million tokens (cache hit)
57+
description: `DeepSeek-V4-Flash is the fast, cost-efficient Mixture-of-Experts model in the DeepSeek-V4 Preview series with 284B total parameters and 13B activated parameters. Supports a 1M-token context window with up to 384K output tokens, thinking mode (default), JSON output, tool calls, chat prefix completion (beta), and FIM completion (beta, non-thinking only). Best for high-volume workloads, chatbots, long document processing, and cost-sensitive applications.`,
58+
},
3559
} as const satisfies Record<string, ModelInfo>
3660

3761
// https://api-docs.deepseek.com/quick_start/parameter_settings

‎packages/types/src/providers/fireworks.ts‎

Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,7 @@ export type FireworksModelId =
55
| "accounts/fireworks/models/kimi-k2-instruct-0905"
66
| "accounts/fireworks/models/kimi-k2-thinking"
77
| "accounts/fireworks/models/kimi-k2p5"
8+
| "accounts/fireworks/models/kimi-k2p6"
89
| "accounts/fireworks/models/minimax-m2"
910
| "accounts/fireworks/models/minimax-m2p1"
1011
| "accounts/fireworks/models/qwen3-235b-a22b-instruct-2507"
@@ -47,6 +48,15 @@ export const fireworksModels = {
4748
description:
4849
"Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters. Trained with the Muon optimizer, Kimi K2 achieves exceptional performance across frontier knowledge, reasoning, and coding tasks while being meticulously optimized for agentic capabilities.",
4950
},
51+
"accounts/fireworks/models/kimi-k2p6": {
52+
maxTokens: 256000,
53+
contextWindow: 262144,
54+
supportsImages: true,
55+
supportsPromptCache: true,
56+
inputPrice: 0.95,
57+
outputPrice: 4.0,
58+
cacheReadsPrice: 0.16,
59+
},
5060
"accounts/fireworks/models/kimi-k2-thinking": {
5161
maxTokens: 16000,
5262
contextWindow: 256000,

‎packages/types/src/providers/moonshot.ts‎

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -66,6 +66,21 @@ export const moonshotModels = {
6666
description:
6767
"Kimi K2.5 is the latest generation of Moonshot AI's Kimi series, featuring improved reasoning capabilities and enhanced performance across diverse tasks.",
6868
},
69+
"kimi-k2.6": {
70+
maxTokens: 16_384,
71+
contextWindow: 262_144,
72+
supportsImages: true,
73+
supportsPromptCache: true,
74+
supportsReasoningBinary: true,
75+
preserveReasoning: true,
76+
inputPrice: 0.95, // $0.95 per million tokens (cache miss)
77+
outputPrice: 4.0, // $4.00 per million tokens
78+
cacheReadsPrice: 0.16, // $0.16 per million tokens (cache hit)
79+
supportsTemperature: false,
80+
defaultTemperature: 0.6,
81+
description:
82+
"Kimi K2.6 is Kimi's most intelligent model to date, featuring stronger and more stable long-horizon code writing capabilities and significantly enhanced autonomous execution capabilities.",
83+
},
6984
} as const satisfies Record<string, ModelInfo>
7085

7186
export const MOONSHOT_DEFAULT_TEMPERATURE = 0.6

‎packages/types/src/providers/openai-codex.ts‎

Lines changed: 16 additions & 16 deletions
Original file line numberDiff line numberDiff line change
@@ -16,14 +16,29 @@ import type { ModelInfo } from "../model.js"
1616

1717
export type OpenAiCodexModelId = keyof typeof openAiCodexModels
1818

19-
export const openAiCodexDefaultModelId: OpenAiCodexModelId = "gpt-5.3-codex"
19+
export const openAiCodexDefaultModelId: OpenAiCodexModelId = "gpt-5.5"
2020

2121
/**
2222
* Models available through the Codex OAuth flow.
2323
* These models are accessible to ChatGPT Plus/Pro subscribers.
2424
* Costs are 0 as they are covered by the subscription.
2525
*/
2626
export const openAiCodexModels = {
27+
"gpt-5.5": {
28+
maxTokens: 128000,
29+
contextWindow: 1000000,
30+
includedTools: ["apply_patch"],
31+
excludedTools: ["apply_diff", "write_to_file"],
32+
supportsImages: true,
33+
supportsPromptCache: true,
34+
supportsReasoningEffort: ["low", "medium", "high", "xhigh"],
35+
reasoningEffort: "medium",
36+
// Subscription-based: no per-token costs
37+
inputPrice: 0,
38+
outputPrice: 0,
39+
supportsTemperature: false,
40+
description: "GPT-5.5: OpenAI's latest frontier model for agentic coding via ChatGPT subscription",
41+
},
2742
"gpt-5.1-codex-max": {
2843
maxTokens: 128000,
2944
contextWindow: 400000,
@@ -172,21 +187,6 @@ export const openAiCodexModels = {
172187
supportsTemperature: false,
173188
description: "GPT-5.1 Codex Mini: Faster version for coding tasks via ChatGPT subscription",
174189
},
175-
"gpt-5.5": {
176-
maxTokens: 128000,
177-
contextWindow: 1_050_000,
178-
includedTools: ["apply_patch"],
179-
excludedTools: ["apply_diff", "write_to_file"],
180-
supportsImages: true,
181-
supportsPromptCache: true,
182-
supportsReasoningEffort: ["none", "low", "medium", "high", "xhigh"],
183-
reasoningEffort: "none",
184-
inputPrice: 0,
185-
outputPrice: 0,
186-
supportsVerbosity: true,
187-
supportsTemperature: false,
188-
description: "GPT-5.5: Most capable model via ChatGPT subscription",
189-
},
190190
"gpt-5.4": {
191191
maxTokens: 128000,
192192
contextWindow: 1_050_000,

0 commit comments

Comments
 (0)