2026-05-29 Digest

Most Impactful Change

• Model: DeepSeek: DeepSeek V3.2 Speciale
• What changed: Removed from the model catalog
• Numbers: Prompt price was $0.287 / 1M tokens, completion price $0.431 / 1M tokens
• Who should care: Teams running high‑volume, cost‑sensitive workloads; loss of this low‑cost option may raise effective inference spend and prompt a search for cheaper substitutes.

Added Models

• Model: StepFun: Step 3.7 Flash

  • What changed: Newly available
  • Numbers: Prompt $0.20 / 1M, completion $1.15 / 1M (context 256k)
  • Who should care: Users needing large‑context generation at moderate cost; suitable for document‑level tasks.

• Model: Anthropic: Claude Opus 4.8 (Fast)

  • What changed: Newly available
  • Numbers: Prompt $10.00 / 1M, completion $50.00 / 1M (context 1M)
  • Who should care: Enterprises prioritizing top‑tier reasoning speed and willing to pay a premium.

• Model: Anthropic: Claude Opus 4.8

  • What changed: Newly available
  • Numbers: Prompt $5.00 / 1M, completion $25.00 / 1M (context 1M)
  • Who should care: Organizations needing high‑quality outputs with slightly lower latency than the Fast variant.

Removed Model (aside from the lead)

• Model: Baidu: Qianfan-OCR-Fast
  • What changed: Removed
  • Numbers: Prompt $0.68 / 1M, completion $2.81 / 1M (context 65k)
  • Who should care: OCR‑focused pipelines that relied on this model’s pricing; may need to adjust OCR service costs.

Price Change

• Model: Z.ai: GLM 4.5 Air
  • What changed: Completion price increased
  • Numbers: Old completion $0.84 / 1M, new completion $0.85 / 1M (prompt unchanged at $0.125 / 1M)
  • Who should care: Developers using this model for completion‑heavy workloads; budget impact is minimal (~1% rise).

Cheapest Models Today (for reference)

1. inclusionAI: Ling-2.6-flash – $0.01 / 1M prompt, $0.03 / 1M completion
2. IBM: Granite 4.0 Micro – $0.017 / 1M prompt, $0.112 / 1M completion
3. Meta: Llama 3.1 8B Instruct – $0.02 / 1M prompt, $0.05 / 1M completion

Total models tracked: 357.

---

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