跪拜 Guibai
← All articles
Developer

OpenCode Go Bundles DeepSeek V4 Flash, Qwen3.8 Max, and GLM-5.2 for $10/Month

By 神奇小汤圆 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

DeepSeek is about to raise API prices significantly, and OpenCode Go arrives as a fixed-cost alternative that bundles multiple frontier Chinese models under one subscription. For developers already using coding agents, it replaces per-token billing anxiety with a predictable $10/month ceiling while keeping access to the same models.

Summary

OpenCode Go is a monthly model subscription built for coding agents. For $5 the first month and $10 thereafter, one API key unlocks DeepSeek V4 Flash, Qwen3.8 Max, GLM-5.2, Kimi K3, MiniMax M3, and Kimi K2.7 Code. The service is separate from OpenCode itself, which remains free and open-source, and the key can also be used with other agents that support custom providers.

DeepSeek V4 Flash's cache-hit pricing of $0.0028 per million tokens makes it particularly cheap inside agent loops that carry large, repeated contexts. A shared bill showed a single Codex request with 156k–190k input tokens costing only $0.0006–$0.0014. The plan does carry soft caps: $12 per 5 hours, $30 per week, and $60 per month, with per-model quotas that vary. Qwen3.8 Max, for instance, is capped at $15/month equivalent and is better reserved for complex planning tasks.

Setup takes minutes. Install OpenCode via curl or npm, run `/connect` and paste the Go API key, then use `/models` to pick or switch models. The key works across terminal, desktop, and IDE environments, and model switching requires no reconfiguration.

Takeaways
OpenCode Go costs $5 for the first month and $10/month after, providing one API key for multiple models.
Supported models include DeepSeek V4 Flash, Qwen3.8 Max, GLM-5.2, Kimi K3, MiniMax M3, and Kimi K2.7 Code.
DeepSeek V4 Flash's cache-read price of $0.0028 per million tokens keeps agent requests extremely cheap; a 156k–190k input-token request cost $0.0006–$0.0014.
The plan enforces soft usage caps: $12 per 5 hours, $30 per week, and $60 per month, with per-model quotas that differ.
Qwen3.8 Max has a monthly equivalent quota of only $15, making it unsuitable as a default model.
Setup requires installing OpenCode, running `/connect` with the Go API key, and selecting a model via `/models`.
The API key works with other coding agents that support custom providers, such as Codex via opencodex.
DeepSeek V4 Flash uses an OpenAI-compatible interface; Qwen3.8 Max uses Anthropic Messages, so direct OpenCode usage avoids protocol headaches.
Conclusions

OpenCode Go is effectively a model arbitrage play: it buys API access at wholesale rates and resells it as a flat subscription, betting that aggregate usage stays within its caps.

The timing is opportunistic. DeepSeek's announced price hike creates immediate demand for a fixed-cost alternative, and OpenCode Go positions itself as the hedge.

Cache-hit economics make or break the value here. DeepSeek V4 Flash's absurdly low cache-read price is what lets a $10 subscription feel unlimited for coding-agent workloads that recycle context heavily.

The per-model quota system means the subscription is really a DeepSeek V4 Flash plan with occasional access to heavier models, not a uniform all-you-can-eat buffet.

Concepts & terms
OpenCode
An open-source AI coding agent that runs in the terminal, desktop, and IDE. It reads code, modifies files, executes commands, and invokes development tools, similar in form to Claude Code or Codex CLI.
OpenCode Go
A monthly subscription service from OpenCode that provides an API key granting access to multiple frontier models (DeepSeek, Qwen, GLM, Kimi, MiniMax) for use inside coding agents, with soft usage caps.
Cache-hit pricing
A billing model where repeated or previously-seen input tokens are charged at a drastically lower rate. DeepSeek V4 Flash charges $0.0028 per million cached input tokens, making it cheap for agent loops that resend large codebases.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗