AI模型价格周报：Qwen3.5 397B涨价$1.26/M，MiniMax M2.7降价20%

今日最具成本影响力的变化是 Qwen 的补全价格飙升：Qwen3.5 397B A17B 上涨 $+1.26 / 1M tokens。

价格变动（每百万 token 视角）

Qwen: Qwen3.5 397B A17B

Prompt: $0.39 → $0.50 (+$0.11) Completion: $2.34 → $3.60 (+$1.26)

Who should care: Teams running large-scale generation workloads with this model; completion costs now exceed $3.50/M.

Qwen: Qwen3.6 27B

Prompt: $0.60 → $0.32 (−$0.28) Completion: $3.60 → $3.20 (−$0.40)

Who should care: Users of this 27B variant see a 19% prompt and 11% completion cost cut.

DeepSeek: DeepSeek V4 Pro 0423

Prompt: $0.40 → $0.53 (+$0.13) Completion: $0.79 → $1.05 (+$0.26)

Who should care: Applications relying on the Pro tier will see ~33% higher completion spend.

MiniMax: MiniMax M2.7

Prompt: $0.30 → $0.24 (−$0.06) Completion: $1.20 → $0.96 (−$0.24)

Who should care: Cost-sensitive inference pipelines benefit from a 20% completion reduction.

DeepSeek: DeepSeek V4 Flash 0731

Prompt: $0.08 → $0.14 (+$0.06) Completion: $0.18 → $0.28 (+$0.10)

Who should care: Flash users face a 75% prompt and 56% completion increase.

DeepSeek: DeepSeek V4 Flash Latest

Prompt: unchanged at $0.04 Completion: $0.13 → $0.08 (−$0.05)

Who should care: Minor completion savings for latency-critical flash workloads.

DeepSeek: DeepSeek V4 Flash 0423

Prompt: $0.05 → $0.06 (+$0.006) Completion: $0.10 → $0.11 (+$0.011)

Who should care: Negligible impact; monitor only if operating at very high volume.

今日最便宜模型（每百万 token）

• inclusionAI: Ling-2.6-flash – Prompt $0.01, Completion $0.03
• IBM: Granite 4.0 Micro – Prompt $0.017, Completion $0.112
• Mistral: Mistral Nemo – Prompt $0.019, Completion $0.030

Total models tracked: 422.

Originally published at The Token Ledger. Subscribe for the daily digest.