Tencent Cloud Token Plan model library update: GLM-5.2 and DeepSeek V4 OEM added, MiniMax-M2.5 retired — pricing comparison

Library update: DeepSeek V4 OEM arrives

Tencent Cloud Token Plan just refreshed its General model library: DeepSeek-V4-Flash / V4-Pro OEM and GLM-5.2 were added, while MiniMax-M2.5 is now retired. The official "Token Plan Individual package overview" lists the current models and each retirement date (source).

The core of this update is bringing DeepSeek's official release directly into the plan. The previous DeepSeek-V4-Flash was Tencent-hosted; the new "OEM" model IDs are served directly by DeepSeek with no TokenHub SLA. The immediate benefit is access to DeepSeek's latest official release—but you accept DeepSeek's own terms of service.

Currently available models

• Auto (tc-code-latest) smart routing, the default.
• DeepSeek-V4-Flash official OEM (deepseek-v4-flash-202605)
• DeepSeek-V4-Pro OEM (deepseek-v4-pro-202606)
• MiniMax-M2.7, GLM-5, GLM-5.1, GLM-5.2
• Kimi-K2.5 (retiring 2026-08-31)

The pricing comparison dimension: additions and retirements

The key pricing comparison isn't about price changes—the model pool changed: at the same General prices you now get GLM-5.2 and direct DeepSeek access, but MiniMax-M2.5 and several old Hunyuan models are gone. The four General tiers' monthly fees and quotas are unchanged (source):

The pricing comparison conclusion: prices didn't move, but the models you can use did. If you relied on MiniMax-M2.5, switch to MiniMax-M2.7 or Auto; if you've been waiting for DeepSeek's official release, you can now use it directly.

Retirement list

DeepSeek V4 OEM: is it worth switching?

If you already use DeepSeek inside Tencent Cloud Token Plan, switching to the OEM path is essentially a free upgrade and worth it. DeepSeek-V4-Flash official corresponds to DeepSeek's V4-Flash-0731 release; set the model parameter to deepseek-v4-flash-202605, and existing users on the old ID are silently upgraded (source).

But be clear on the boundary: the OEM path is served directly by DeepSeek with no TokenHub SLA, and you must accept DeepSeek's terms of service first. For stability-sensitive workloads, Auto routing or the hosted Zhipu GLM, MiniMax and Kimi models may be safer.

Hy3 GA: the Hy series' hidden update

This update isn't just the General series—Hy also moved Hy3 from preview to GA, with hy3-preview calls auto-routing to Hy3. Tencent says Hy3 GA was hardened from real workload feedback across Tencent Yuanbao, WorkBuddy and others, improving coding agent, long-context and complex task execution (source).

The Hy series is cheaper than General (Lite ¥28, Standard ¥78, Pro ¥238, Max ¥468) but its pool is Hy3 only. The pricing comparison conclusion: if you can live with Hunyuan Hy3 alone, Hy is more economical; pick General when you need multi-model switching.

Integration and boundaries

Tencent Cloud Token Plan's integration is unchanged: General and Hy share one API key and deduct from the matching plan pool by Model ID. Key setup points (source):

• OpenAI-compatible Base URL: https://api.lkeap.cloud.tencent.com/plan/v3
• Anthropic-compatible: https://api.lkeap.cloud.tencent.com/plan/anthropic
• Supports Claude Code, Cursor, OpenCode, Cline, CodeBuddy, OpenClaw and more.
• Interactive coding/agent tools only—no account sharing or backend API calls.

Things to watch

1. MiniMax-M2.5 is retired (Aug 7). Switch to MiniMax-M2.7 or Auto now, or calls will fail.

2. Kimi-K2.5 retires Aug 31. Migrate your workflow to Auto, GLM, or MiniMax series ahead of time.

3. DeepSeek OEM has no SLA. It's served directly by DeepSeek without a TokenHub guarantee—read DeepSeek's terms of service first.

4. Prices are unchanged, but the model pool is dynamic. Tencent states the library may shift with performance, compliance, and licensing, so don't treat any single model as permanent.