DeepSeek for pocket change: pay $10, code like you paid $100

OpenCode, the open-source AI coding agent, has kicked off phase four of a promo with a very honest name, Operation Cheepseek. Its OpenCode Go plan, which costs $10 a month, now includes $100 of usage for the Chinese model DeepSeek V4.1 Flash. The announcement went out on the evening of October 30 and passed 700,000 views within a day.

The deal has been climbing in steps. Go first came with $30 of DeepSeek usage, then $60, then $80, and now $100. Nobody hands you cash, of course: it is an allowance measured against regular token prices. On the $40 Go Plus plan the DeepSeek V4.1 Flash allowance doubles to $200.

OpenCode did the math on what that means in practice. By its estimate, under a typical coding agent workload $100 covers roughly 216,000 requests a month. There are guardrails: any five hour window can use at most 20% of the monthly allowance, and any week at most half. Token prices also move. During weekday peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC) they double, while the rest of the time, weekends included, off-peak pricing applies and the allowance lasts longer.

Setup takes a couple of commands. Subscribe to Go in the OpenCode console, copy your key, run /connect in the agent, pick OpenCode Go, paste the key and choose a model with /models. DeepSeek has one catch: by default only servers in Europe and the US are allowed, so you need to switch on the Global region in the privacy settings. The key is not limited to OpenCode itself either, the company has validated it with Claude Code, Codex, Hermes and Pi. Under the service terms, DeepSeek requests are not used for training and are not retained.

DeepSeek V4.1 Flash came out in September and handles a context of about a million tokens. The same plan also includes other models such as Kimi K3.1 Pro, Qwen3.8, GLM-5.3 and MiMo, though their allowances are much smaller. The same day OpenCode also learned to work with a Claude subscription, as we reported.

Right now this looks like the cheapest way to put an AI agent to work on your code.