AI 模型变更日志 · 2026-06-26

2026-06-26 当日记录 50 个事件。

新增模型 (9)

Mistral New model: Mistral Medium 3.5 128B
vercel New model: GLM 5.2 Fast
neuralwatt New model: GLM 5.2 short
llmgateway New model: Mistral OCR
NVIDIA New model: Phi 4 Multimodal
evroc New model: Qwen3 Reranker 4B
evroc New model: Whisper Large v3 Turbo
evroc New model: roc
openrouter New model: Fugu Ultra

价格变化 (12)

Alibaba (Qwen)Price update: Qwen3 30B A3B Instruct 2507 now $0.09/$0.3 (was $0.04815/$0.19305)
evroc Price update: KB Whisper now $0.0023/$0.0023 (was $0.00236/$0.00236)
evroc Price update: Llama-3.3-70B-Instruct now $1.15/$1.15 (was $1.18/$1.18)
evroc Price update: E5 Multi-Lingual Large Embeddings 0.6B now $0.114/$0.114 (was $0.12/$0.12)
llmgateway Price update: GPT OSS 120B now $0.05/$0.25 (was $0.15/$0.75)
llmgateway Price update: GPT OSS 20B now $0.04/$0.15 (was $0.1/$0.5)
Meta Price update: Llama 4 Maverick 17B Instruct now $0.27/$0.85 (was $0.24/$0.97)
Meta Price update: Llama 4 Scout 17B Instruct now $0.18/$0.59 (was $0.17/$0.66)
MiniMax Price update: MiniMax-M3 now $0.3/$1.2 (was $0.6/$2.4)
Mistral Price update: Voxtral Small 24B 2507 now $0.0023/$0.0023 (was $0.00236/$0.00236)
OpenAI Price update: Whisper Large v3 now $0.0023/$0.0023 (was $0.00236/$0.00236)
StepFun Price update: Step 3.7 Flash now $0.19/$1.13 (was $0.2/$1.15)

上下文变化 (5)

DeepSeek Context window: DeepSeek V3.1 Terminus now 128,000 (was 163,840)
evroc Context window: Llama-3.3-70B-Instruct now 128,000 (was 131,072)
MiniMax Context window: MiniMax-M3 now 1,000,000 (was 512,000)
Moonshot AI Context window: MoonshotAI Kimi Latest now 256,000 (was 262,144)
openrouter Context window: Nano Banana 2 (Gemini 3.1 Flash Image) now 131,072 (was 65,536)

输出上限变化 (6)

DeepSeek Max output: DeepSeek V3.1 Terminus updated
evroc Max output: Llama-3.3-70B-Instruct updated
Mistral Max output: Mistral: Mistral 7B Instruct v0.1 updated
Moonshot AI Max output: MoonshotAI Kimi Latest updated
openrouter Max output: Nano Banana 2 (Gemini 3.1 Flash Image) updated
openrouter Max output: Nex-N2-Pro (free) updated

能力字段更新 (18)

DeepSeek Capabilities updated: DeepSeek V3.2 Thinking
evroc Capabilities updated: Llama-3.3-70B-Instruct
jiekou Capabilities updated: gemini-2.5-flash-lite-preview-06-17
jiekou Capabilities updated: gemini-2.5-flash-preview-05-20
jiekou Capabilities updated: gemini-2.5-pro-preview-06-05
Mistral Capabilities updated: Mistral: Mistral 7B Instruct v0.1
openrouter Capabilities updated: Nex-N2-Pro (free)
StepFun Capabilities updated: Step 3.7 Flash
synthetic Capabilities updated: Nemotron 3 Super 120B A12B
venice Capabilities updated: Qwen 3.7 Max
vercel Capabilities updated: Qwen 3.6 Max Preview
vercel Capabilities updated: Devstral Small 2
vercel Capabilities updated: Interfaze Beta
vercel Capabilities updated: Nova 2 Lite
vercel Capabilities updated: Qwen3 VL Instruct
vercel Capabilities updated: Seed 1.6
xAI Capabilities updated: Grok 3 Beta
xAI Capabilities updated: Grok 3 Mini Beta

Pricing and capabilities are refreshed daily and reconciled against each provider's official documentation. Always verify critical production decisions with the provider directly.