Token Ledger Digest – 2026-09-05

Biggest cost impact

Qwen: Qwen3.6 27B – completion price fell from $3.60 to $2.00 per M tokens (‑$1.60/M); prompt price dropped from $0.60 to $0.30 per M tokens (‑$0.30/M).

Who should care: Teams running long‑form generation or chat workloads on this model see ~35% lower variable cost.

Other notable price changes

• MoonshotAI Kimi Latest – completion ↓$1.25/M (from $14.00 to $12.75/M); prompt ↑$0.05/M (from $2.50 to $2.55/M).
• DeepSeek V4 Pro 0423 – prompt ↓$0.15/M; completion ↓$0.30/M.
• Qwen3.5‑35B‑A3B – prompt ↓$0.17/M; completion ↓$0.50/M.
• DeepSeek V4 Flash Latest – completion ↓$0.06/M (tiny prompt change).
• Qwen3 235B A22B Instruct 2507 – completion ↑$0.20/M; prompt ↑$0.0025/M.
• DeepSeek V4 Pro 0813 – prompt ↑$0.005/M; completion ↑$0.016/M.
• DeepSeek V4 Flash 0423 – prompt ↓$0.003/M; completion ↓$0.007/M.

New models added

• OpenAI: GPT‑6 Astra – prompt $10/M, completion $50/M (context 1.05M).
• OpenAI: GPT‑6 Astra (batch) – prompt $5/M, completion $25/M.
• OpenAI: GPT‑6 Astra Pro – same as Astra; batch version halves both rates.
• inclusionAI: Ling 3.0 Flash Sante (free) – $0 prompt & completion.
• Qwen: Qwen3.8 Max (0902) – prompt $2/M, completion $6/M (context 1M).

Models removed

• Qwen: Qwen3.8 Max (same pricing as the new 0902 version).
• IBM: Granite 4.1 8B – prompt $0.05/M, completion $0.10/M.

Cheapest models today (per M tokens)

1. IBM Granite 4.0 Micro – prompt $0.017, completion $0.112
2. Mistral Nemo – prompt $0.019, completion $0.030
3. inclusionAI Ling 3.0 Flash – prompt $0.021, completion $0.063

Total models tracked: 431.

---

All prices shown in USD per 1 M tokens.

---

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