Token Ledger Digest – 2026-08-26

Most cost‑impacting change

• Qwen: Qwen3.5 397B A17B – Prompt price fell from $0.50 to $0.39 per 1M tokens (‑$0.11/1M); completion price fell from $3.60 to $2.34 per 1M tokens (‑$1.26/1M). Who should care: Teams running large mixture‑of‑experts workloads see per‑token cost drop ~35%.

Other price changes

Model	Prompt ($/1M) Old → New	Completion ($/1M) Old → New
DeepSeek V4 Flash Latest	0.035 → 0.030	0.10 → 0.075
DeepSeek V4 Flash 0731	0.14 → 0.06	0.28 → 0.12
MoonshotAI Kimi Latest	2.80 → 2.55	14.0 → 12.75
Qwen3.6 27B	0.32 → 0.60	3.20 → 3.60
DeepSeek V4 Pro 0423	0.790 → 0.870	1.581 → 1.740
Gemma 4 31B	0.10 → 0.09	0.34 → 0.34 (unchanged)
Llama 4 Scout	0.10 → 0.11	0.30 → 0.34
Qwen2.5 VL 72B Instruct	0.80 → 0.25	1.00 → 0.75
Llama 3.3 70B Instruct	0.10 → 0.71	0.32 → 0.71

Removed models

• Qwen: Qwen Plus 0728 (thinking) – Prompt $0.26/1M, Completion $0.78/1M, 1M‑token context.
• Google: Gemma 3n 4B – Prompt $0.06/1M, Completion $0.12/1M, 32 K‑token context. Who should care: Users of these models must switch to alternatives; no further pricing data will be available.

Total models tracked: 417. No additions today.

---

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