Token Ledger Digest - 2026-09-20

Lead change – biggest cost impact

• Model: DeepSeek: DeepSeek V4 Pro 0423
• What changed: Prompt price fell from $1.60 / 1M to $0.42 / 1M; completion price fell from $3.20 / 1M to $0.84 / 1M.
• Numbers: Old total ≈ $4.80 / 1M tokens; new total ≈ $1.26 / 1M tokens (‑$3.53 / 1M).
• Who should care: Teams running large‑scale generation or fine‑tuning with DeepSeek V4 Pro now see a ~73% reduction in token cost.

Added models

• PrismML: Ternary Bonsai 2 27B – Prompt $0.075 / 1M, Completion $0.50 / 1M, Context 262 k.
• Z.ai: GLM 5.3 FlashX – Prompt $0.37 / 1M, Completion $1.25 / 1M, Context 1,048 k.

Removed model

• DeepSeek: DeepSeek V4 Flash 0731 (free) – No longer available; the free tier is gone.

Other notable price shifts

• Qwen: Qwen3.8 27B – Prompt ↑$0.21→$0.42 / 1M (+$0.21), Completion ↑$2.55→$3.00 / 1M (+$0.45).
• MoonshotAI: Kimi K3 / Kimi Latest – Prompt ↓$2.10→$1.70 / 1M, Completion ↓$10.95→$8.50 / 1M (‑$2.45 / 1M each).
• Z.ai: GLM 5.2 – Prompt ↑$0.56→$0.65 / 1M, Completion ↑$1.80→$2.04 / 1M.
• NVIDIA: Nemotron 3 Ultra – Prompt ↓$0.63→$0.60 / 1M, Completion ↓$3.13→$2.40 / 1M.

Cheapest models today (per‑million rates)

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

Total models tracked: 446.

No hype, just the numbers.

---

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