Token Ledger Digest – 2026-09-23
Most cost‑impacting change: Anthropic’s Claude Opus Latest saw its prompt price drop from $5.00 to $4.00 per 1M tokens and completion from $25.00 to $20.00 per 1M tokens, a $6/M total reduction. Cost‑focused teams using Opus should revisit forecasted spend.
Price Changes
- MoonshotAI Kimi Latest: prompt ≈$1.50 → $1.50/M; completion rose $7.50 → $10.76/M (+$3.26/M). Users needing long completions will see higher bills.
- DeepSeek Pro Latest: prompt $0.62 → $0.40/M; completion $2.88 → $4.30/M (+$1.42/M completion, –$0.22/M prompt). Net cost up for completion‑heavy workloads.
- DeepSeek V4 Pro 0813: prompt $0.62 → $0.50/M; completion $2.88 → $1.50/M (–$1.38/M). Savings for both prompt and completion.
- OpenAI GPT Luna Latest: prompt $0.20 → $0.10/M; completion $1.20 → $0.50/M (–$0.70/M). Benefits latency‑sensitive apps.
- Z.ai GLM 5.3 Flash (batch): prompt $0.075 → $0.06/M; completion $0.25 → $0.20/M (–$0.065/M). Minor edge for batch pipelines.
Added Models (selected)
- Upstage: Solar Mini 4 – prompt $0.05/M, completion $0.20/M, 524k context. Cheapest new entry; ideal for low‑cost, long‑context tasks.
- DeepSeek V4.1 Flash (batch) – prompt $0.11/M, completion $0.34/M, 1,048,576 context. Good batch‑oriented flash alternative.
- OpenAI: GPT‑6 Luna Pro – prompt $0.10/M, completion $0.50/M, 1.05M context. Balanced price/performance for large‑scale generation.
Removed Models
- DeepSeek V4 Flash Vision Exp (batch)
- DeepSeek V4 Pro 0813 (batch)
- Meta Muse Glimmer 30B (batch)
- DeepSeek V4 Flash 0731 (batch)
- Z.ai GLM 5.2 (batch)
These batch variants are no longer offered; users should migrate to non‑batch versions or alternatives.
Cheapest Models Today
- IBM Granite 4.0 Micro – $0.017/M prompt, $0.112/M completion
- OpenAI gpt‑oss‑20b – $0.018/M prompt, $0.09/M completion
- Mistral Nemo – $0.019/M prompt, $0.03/M completion
Total models tracked: 455.
Originally published at The Token Ledger. Subscribe for the daily digest.
Top comments (0)