Token Ledger Digest – 2026-09-24

Most cost‑impacting change

• DeepSeek: DeepSeek Pro Latest – completion price fell from $4.30 / 1M to $1.1682 / 1M (‑73%). Who should care: Teams running long‑form generation or chat workloads on this model will see per‑token costs drop sharply; revisit token budgets and consider shifting volume here.

Other notable price changes

• DeepSeek: Flash Latest – completion rose from $0.50 / 1M to $1.00 / 1M (+100%).
• DeepSeek: V4 Flash 0731 – completion dropped from $0.64 / 1M to $0.32 / 1M (‑50%).
• MiniMax: M2 – prompt up $0.255 → $0.30 / 1M (+18%); completion up $1.02 → $1.20 / 1M (+18%).
• Qwen: 30B A3B Instruct 2507 – prompt up $0.048 → $0.10 / 1M (+108%); completion up $0.193 → $0.30 / 1M (+55%).

Added models (selected)

• Space Bunny Alpha – zero‑cost prompt & completion (free tier). Good for prototyping or cost‑sensitive experiments.
• OpenAI: gpt‑oss‑120b (batch) – prompt $0.0296 / 1M, completion $0.136 / 1M; attractive for high‑volume batch jobs.
• AionLabs: Aion 3.5 Mini – prompt $0.70 / 1M, completion $1.40 / 1M; low‑cost option for smaller context windows.

Removed models

• Four models dropped, including the free inclusionAI: Ling 3.0 Flash VL and two Nex AGI variants. Users relying on these should migrate to alternatives listed above.

Cheapest models today (per‑million tokens)

1. IBM: Granite 4.0 Micro – $0.017 / 1M prompt, $0.112 / 1M completion
2. OpenAI: gpt‑oss‑20b – $0.018 / 1M prompt, $0.090 / 1M completion
3. Mistral: Mistral Nemo – $0.019 / 1M prompt, $0.030 / 1M completion

Total models tracked: 458.

Developers should adjust cost models for the DeepSeek Pro Latest drop, watch for rising Flash‑series costs, and evaluate the new zero‑cost and batch offerings for experimental or high‑throughput workloads.

---

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