Token Ledger Digest – 2026-05-31
Cost‑impacting change
Qwen: Qwen3 235B A22B Thinking 2507
- Prompt: fell from $0.1495/1M to $0.10/1M (‑33%).
- Completion: fell from $1.495/1M to $0.10/1M (‑93%).
- Who should care: Teams running large‑scale generation workloads where output token cost dominates; this cut reduces per‑million completion expense by ~$1.40.
Other price changes
MiniMax: MiniMax M2.7
- Prompt: $0.279/1M → $0.26/1M (‑7%). Completion unchanged at $1.20/1M.
- Relevant for users prioritizing prompt‑heavy tasks.
OpenAI: gpt-oss-20b
- Prompt: $0.03/1M → $0.029/1M (‑3%). Completion unchanged at $0.14/1M.
- Minor saving for latency‑sensitive apps using this model.
Model removals (6)
| Model | Context | Prompt ($/1M) | Completion ($/1M) |
|---|---|---|---|
| MiniMax: MiniMax M2.5 (free) | 204,800 | 0.00 | 0.00 |
| Upstage: Solar Pro 3 | 128,000 | 0.15 | 0.60 |
| Baidu: ERNIE 4.5 21B A3B Thinking | 131,072 | 0.07 | 0.28 |
| Baidu: ERNIE 4.5 21B A3B | 131,072 | 0.07 | 0.28 |
| AlfredPros: CodeLLaMa 7B Instruct Solidity | 4,096 | 0.80 | 1.20 |
| Mistral: Mistral 7B Instruct v0.1 | 4,096 | 0.11 | 0.19 |
Developers relying on any of these models must migrate to alternatives; none of the removed entries offered a free tier except the MiniMax M2.5 variant.
Cheapest models today (per‑million)
- inclusionAI: Ling-2.6-flash – Prompt $0.01, Completion $0.03
- IBM: Granite 4.0 Micro – Prompt $0.017, Completion $0.112
- Meta: Llama 3.1 8B Instruct – Prompt $0.02, Completion $0.05
Total models tracked: 350. No other meaningful changes recorded.
Originally published at The Token Ledger. Subscribe for the daily digest.
Top comments (0)