Token Ledger Digest – 2026-07-25

Lead change – biggest cost impact

• Model: NVIDIA: Nemotron 3 Ultra
• What changed: Prompt price rose from $0.50 to $0.60 per 1M tokens; completion price rose from $2.20 to $3.60 per 1M tokens.
• Numbers: +$0.10 / 1M prompt, +$1.40 / 1M completion → +$1.50 / 1M total.
• Who should care: Teams running large‑scale inference with this model will see a 64% jump in completion cost; consider alternatives or renegotiate usage.

Other price changes (Δ per 1M tokens = prompt Δ + completion Δ)

Model	Δ Prompt ($/1M)	Δ Completion ($/1M)	Δ Total ($/1M)	Note
Z.ai: GLM 5.2	–0.004	–0.013	–0.017	Small cost cut
MoonshotAI: Kimi K2.7 Code	+0.070	+0.010	+0.080	Slight increase
DeepSeek: V4 Flash	–0.004	–0.008	–0.012	Minor reduction
MoonshotAI: Kimi K2.6	–0.038	–0.700	–0.738	Notable completion drop
Google: Gemma 4 31B	+0.040	+0.050	+0.090	Modest rise
Arcee AI: Trinity Large Thinking	–0.030	+0.050	+0.020	Mixed
MiniMax: M2.7	–0.050	–0.200	–0.250	Meaningful cut
NVIDIA: Nemotron 3 Super	+0.005	–0.050	–0.045	Slight net decrease
Qwen: Qwen3 Next 80B A3B Instruct	–0.050	–0.100	–0.150	Notable drop
OpenAI: gpt-oss-20b	+0.001	–0.010	–0.009	Tiny change
Qwen: Qwen3 Coder 480B A35B	–0.080	–0.550	–0.630	Large completion cut

Added models

Model	Prompt ($/1M)	Completion ($/1M)	Context
Anthropic: Claude Opus 5 (Fast)	$10.00	$50.00	1M tokens
Anthropic: Claude Opus 5	$5.00	$25.00	1M tokens

Who should care:

• Cost‑sensitive projects should review the Nemotron 3 Ultra increase and consider the newly added Claude Opus 5 variants if higher quality justifies the premium.
• Teams using MoonshotAI Kimi K2.6, Qwen Coder, or MiniMax M2.7 can expect lower completion expenses.
• General users can reference the cheapest‑today list (inclusionAI Ling‑2.6‑flash, IBM Granite 4.0 Micro, Mistral Nemo) for ultra‑low‑cost options.

Total models tracked: 345.

---

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