Token Ledger Digest – 2026-09-17

Most cost‑impacting change: Z.ai’s GLM 5.2 saw its prompt price rise from $0.60 to $1.40 per 1M tokens and completion from $2.00 to $4.40 per 1M tokens – a +$3.20/M increase. Developers running latency‑critical or high‑volume workloads on this model should re‑evaluate cost budgets or consider alternatives.

Added models

Model	Prompt ($/1M)	Completion ($/1M)	Who should care
stealth/union‑alpha	0.00	0.00	Experimental users seeking zero‑cost access.
deepseek/deepseek‑pro‑latest	0.66	1.98	Teams needing strong reasoning at moderate price.
deepseek/deepseek‑flash‑latest	0.15	0.60	Cost‑sensitive apps requiring fast responses.
z‑ai/glm‑5.2:free	0.00	0.00	Anyone wanting a free, medium‑context model.

Removed models

Model	Prompt ($/1M)	Completion ($/1M)	Who should care
thinkingmachines/inkling‑small:batch	0.50	1.20	Batch‑oriented users losing a cheap option.
google/gemma‑4‑31b‑it:batch	0.39	0.97	Users of Google’s batch‑tuned Gemma.
openai/gpt‑oss‑20b:batch	0.05	0.20	Low‑cost batch workloads now unavailable.
google/gemini‑2.5‑pro‑preview‑05‑06	1.25	10.00	High‑preview Gemini users losing access.
openai/gpt‑4‑turbo‑preview	10.00	30.00	Anyone relying on the preview Turbo model.

Notable price changes (top 5 by absolute impact)

Model	Old Prompt ($/1M)	New Prompt ($/1M)	Old Comp. ($/1M)	New Comp. ($/1M)	Net Δ ($/1M)	Who should care
z‑ai/glm‑5.2	0.60 → 1.40	+0.80	2.00 → 4.40	+2.40	+3.20	High‑volume GLM 5.2 users.
moonshotai/kimi‑k3	2.65 → 3.00	+0.35	13.28 → 15.00	+1.72	+2.07	Kimi‑K3 adopters seeing higher spend.
meta/muse‑glimmer‑30b	0.35 → 0.30	–0.05	1.50 → 1.10	–0.40	–0.45	Cost‑conscious Muse users benefit.
z‑ai/glm‑5.3	1.09 → 1.40	+0.31	3.43 → 4.40	+0.97	+1.28	GLM 5.3 workloads now pricier.
deepseek/deepseek‑v4‑pro‑0813	0.58 → 0.66	+0.08	1.73 → 1.98	+0.25	+0.33	DeepSeek V4 Pro users face modest rise.

All other price changes were smaller than $0.33/M.

Total models tracked: 444.

---

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