The Token Ledger – 2026-09-21

Lead change: DeepSeek V4 Pro 0423 saw the largest cost jump, with prompt pricing more than doubling and completion rising ~125%. Developers relying on this model for long‑form generation should budget for ~+$1.58 per M tokens.

Price changes

• DeepSeek Pro Latest – prompt ↑0.5636→0.5808 $/M, completion ↑1.6909→1.7424 $/M. Affected: users of the latest DeepSeek Pro tier.
• DeepSeek Flash Latest – prompt ↓0.13→0.12 $/M, completion ↓0.52→0.48 $/M. Affected: cost‑sensitive Flash workloads.
• IBM Granite 4.2 8B – prompt ↑0.06→0.10 $/M, completion ↓0.25→0.15 $/M. Affected: IBM Granite adopters seeing mixed cost impact.
• DeepSeek V4 Flash Vision Exp – prompt ↑0.2156→0.22 $/M, completion ↑0.6468→0.66 $/M. Affected: vision‑enabled Flash users.
• Z.ai GLM Latest – prompt ↓0.8442→0.7735 $/M, completion ↓2.6532→2.431 $/M. Affected: GLM Latest customers benefiting from lower rates.
• Z.ai GLM 5.3 – prompt ↑0.896→0.91 $/M, completion ↑2.816→2.86 $/M. Affected: GLM 5.3 users facing slight cost rise.
• Qwen Qwen3.8 27B – prompt ↓0.42→0.20 $/M, completion ↓3.0→2.5 $/M. Affected: large‑model Qwen users seeing notable savings.
• DeepSeek V4 Pro 0813 – prompt ↓0.66→0.5808 $/M, completion ↓1.98→1.7424 $/M. Affected: users of the 0813 Pro variant.
• DeepSeek V4 Flash Latest (tilde) – prompt unchanged 0.04 $/M, completion ↑0.08→0.16 $/M. Affected: Flash Latest users with higher completion costs.
• DeepSeek V4 Flash 0731 – same as above. Affected: identical impact for the 0731 snapshot.
• Qwen Qwen3.6 35B A3B – prompt ↑0.10→0.15 $/M, completion ↑0.90→1.00 $/M. Affected: Qwen 3.6 users experiencing moderate cost increase.
• DeepSeek V4 Pro 0423 – prompt ↑0.4223→0.9481 $/M, completion ↑0.8446→1.8963 $/M. Affected: all Pro 0423 deployments (see lead).
• DeepSeek V4 Flash 0423 – prompt ↑0.0364→0.0570 $/M, completion ↑0.0728→0.1140 $/M. Affected: Flash 0423 users.
• Qwen Qwen3 VL 30B A3B Instruct – prompt ↓0.20→0.13 $/M, completion ↓0.70→0.52 $/M. Affected: multimodal Qwen VL users.
• Meta Llama 4 Maverick – prompt ↑0.1875→0.20 $/M, completion ↑0.6525→0.80 $/M. Affected: Llama 4 Maverick adopters.

Cheapest models today (per‑M token rates):

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

No model additions or removals today. Total models tracked: 446.

---

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