Token Ledger Digest – 2026-09-11
Biggest cost shift: MoonshotAI: Kimi K3 – prompt price fell from $3.00 to $2.34 / 1M tokens (‑0.66) and completion from $15.00 to $11.70 / 1M tokens (‑3.30). Total ‑$3.96 / 1M. Who should care: Teams running long‑form generation or chat workloads on Kimi K3 see ~26% lower inference cost.
Added models
- Sakana: Fugu Ultra v2 – 1M‑token context, prompt $5.00 / 1M, completion $30.00 / 1M. Who should care: Users needing very long context with moderate‑priced generation.
- Sakana: Fugu Max – 1M‑token context, prompt $2.00 / 1M, completion $6.00 / 1M. Who should care: Cost‑sensitive applications that still require large context windows.
- inclusionAI: Ling 3.0 Flash VL (free) – 262k‑token context, prompt $0.00, completion $0.00 / 1M. Who should care: Prototyping or education where zero cost is critical.
Other price changes (Δ per 1M tokens)
| Model | Prompt Δ | Completion Δ | Total Δ |
|---|---|---|---|
| Z.ai: GLM Latest | –0.0375 | –0.1023 | –0.1398 |
| DeepSeek: V4 Pro 0813 | –0.3894 | –1.1682 | –1.5576 |
| Upstage: Solar Pro 4 | +0.0600 | +0.2400 | +0.3000 |
| MoonshotAI: Kimi Latest | –0.0600 | –0.3000 | –0.3600 |
| Google: Gemma 4 26B A4B | –0.0280 | –0.1200 | –0.1480 |
| Qwen: Qwen3 30B A3B Instruct 2507 | +0.0419 | +0.1070 | +0.1489 |
| DeepSeek: V3 0324 | –0.0400 | –0.1400 | –0.5400 |
Who should care: Developers monitoring budget impacts—note the notable price rise for Upstage Solar Pro 4 (+$0.30 / 1M) and Qwen3 (+$0.15 / 1M), while most models saw reductions.
Cheapest models today (per 1M tokens)
- IBM: Granite 4.0 Micro – prompt $0.017, completion $0.112
- Mistral: Mistral Nemo – prompt $0.019, completion $0.030
- inclusionAI: Ling 3.0 Flash – prompt $0.021, completion $0.063
Total models tracked: 439. No removals reported.
Originally published at The Token Ledger. Subscribe for the daily digest.
Top comments (0)