Token Ledger Digest – 2026-08-12

Price Changes (most cost‑impacting first)

Z.ai: GLM 5.2

• Prompt: $0.76 → $0.50 /1M tokens (‑$0.26)
• Completion: $2.42 → $3.15 /1M tokens (+$0.73) Who should care: Teams running high‑volume completion workloads will see per‑million costs rise ~30%; prompt‑heavy flows benefit from a 34% drop.

Qwen: Qwen3 VL 235B A22B Instruct

• Prompt: $0.21 → $0.26 /1M (+$0.05)
• Completion: $1.90 → $1.04 /1M (‑$0.86) Who should care: Vision‑language apps that generate long outputs will cut completion spend by ~45%; slight prompt increase is negligible.

DeepSeek: DeepSeek V3.1 Terminus

• Prompt: unchanged $0.27 /1M
• Completion: $1.00 → $0.95 /1M (‑$0.05) Who should care: Minor savings for completion‑heavy use cases.

OpenAI: gpt-oss-120b

• Prompt: $0.037 → $0.030 /1M (‑$0.007)
• Completion: unchanged $0.17 /1M Who should care: Developers using this model for prompt‑intensive tasks save <1% per million tokens.

Added Models

LiquidAI: LFM2.5-2.6B (free)

• Prompt/Completion: $0 /1M, context 128k Who should care: Prototyping or education projects needing zero‑cost inference.

NVIDIA: Nemotron 3.5 Lightning

• Prompt: $0.10 /1M, Completion: $0.25 /1M, context 262k Who should care: Users seeking a low‑cost, mid‑size model with decent throughput.

NVIDIA: Nemotron 3.5 Lightning (free)

• Prompt/Completion: $0 /1M, context 1M Who should care: Applications requiring very long context without cost.

ByteDance Seed: Seed-2.0-Code

• Prompt: $0.50 /1M, Completion: $3.00 /1M, context 262k Who should care: Code‑generation workloads where completion cost is acceptable for higher quality.

Total models tracked: 406.

---

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