Token Ledger Digest – 2026-09-01

Lead change – biggest cost impact

• DeepSeek: DeepSeek V4 Pro 0423 – Prompt price rose from $0.417 / 1M to $1.60 / 1M (+$1.18 / 1M); completion price rose from $0.835 / 1M to $3.20 / 1M (+$2.37 / 1M). Who should care: Teams running high‑volume inference on V4 Pro will see per‑token costs more than triple; budget forecasts need revision.

Other notable price changes

• DeepSeek V4 Flash Latest – Prompt +$0.02 / 1M ($0.03→$0.05); completion unchanged at $0.16 / 1M.
• Z.ai: GLM Latest – Prompt ‑$0.0175 / 1M ($1.1875→$1.17); completion ‑$0.22 / 1M ($4.18→$3.96).
• MoonshotAI: Kimi K2.5 – Prompt ‑$0.15 / 1M ($0.60→$0.45); completion ‑$0.75 / 1M ($3.00→$2.25).
• Meta: Llama 4 Maverick – Completion ‑$0.104 / 1M ($0.80→$0.696); prompt steady at $0.20 / 1M.
• Meta: Llama 4 Scout – Prompt ‑$0.01 / 1M ($0.11→$0.10); completion ‑$0.04 / 1M ($0.34→$0.30).
• Magnum v4 72B – Prompt ‑$0.50 / 1M ($3.00→$2.50); completion unchanged at $5.00 / 1M.
• Mancer: Weaver (alpha) – Prompt ‑$0.10 / 1M ($0.50→$0.40); completion unchanged at $0.75 / 1M.
• Qwen: Qwen3 Next 80B A3B Instruct – Prompt +$0.01 / 1M ($0.09→$0.10); completion steady at $1.10 / 1M.
• DeepSeek V4 Flash 0423 – Prompt +$0.0014 / 1M ($0.0795→$0.0809); completion +$0.0028 / 1M ($0.1590→$0.1618).

Additions / removals

• Added: 30 new models, including IBM Granite 4.2 8B (prompt $0.10 / 1M, completion $0.15 / 1M) and a suite of OpenAI GPT‑5.x batch variants.
• Removed: 7 models, notably the high‑cost Anthropic Claude Opus 4.7 Fast (prompt $30 / 1M, completion $150 / 1M) and several Mistral batch lines.

Cheapest models today (per‑million‑token rates)

1. IBM Granite 4.0 Micro – Prompt $0.017 / 1M, Completion $0.112 / 1M
2. Mistral Nemo – Prompt $0.019 / 1M, Completion $0.030 / 1M
3. Ling‑3.0‑flash – Prompt $0.021 / 1M, Completion $0.063 / 1M

Total models tracked: 419.

---

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