OpenCode DeepSeek V4.1 Flash: Free vs $0.30/$1.20 vs Go Off-Peak — Don’t Mix These Three

If you only remember one thing: OpenCode DeepSeek V4.1 Flash is not one price. It is three (sometimes four) bills that share a model nickname. Mix them and you will “save money” on a spreadsheet that does not match your Console statement.

I checked the public pages again on 2026-10-23 (Asia/Shanghai). The Zen / Console pricing table, the DeepSeek V4.1 Flash data page, and the Go subscription page still disagree unless you read the row labels carefully. That is the whole story.

OpenCode DeepSeek V4.1 Flash: three price paths (do not mix)

Here is the comparison I keep next to the terminal:

Path	How you pay	Input / 1M	Output / 1M	Monthly quota note
—	—	—	—	—
① Timed Free	Zen “Free models” row / ID like deepseek-v4.1-flash-free	$0	$0	Rate-limited; exact RPM not published
② Zen / Console pay-as-you-go	Top up balance, billed per token	$0.30	$1.20	Cached read $0.006 / 1M
③ Go Peak	Go $10 or Go Plus $40 subscription	$0.30	$1.20	Go $80 / Plus $160 on this model
③ Go Off-Peak	Same subscription, off-peak hours	$0.15	$0.60	Same monthly $ quotas

Cached read on Go follows the same half-price pattern Off-Peak ($0.003 vs $0.006). The data page’s list price matches Zen and Go Peak — it is not a promise that Flash is permanently free.

OpenCode DeepSeek V4.1 Flash Free is a timed entry. When the Free row disappears from Zen, that path is gone. The paid rows stay.

Client is free; models are not all free

OpenCode’s FAQ still says the client is 100% free to use. Connecting other providers can cost money. That sentence does not mean every model ID is free forever.

Go Peak hours (UTC) and Beijing time

For OpenCode DeepSeek V4.1 Flash inside Go, Peak is:

• UTC Monday–Friday 01:00–04:00
• UTC Monday–Friday 06:00–10:00

Everything else — including weekends — is Off-Peak.

In Beijing time (UTC+8) that is roughly:

• 09:00–12:00 weekday mornings
• 14:00–18:00 weekday afternoons

If you run a long job at 10:30 Beijing on a Tuesday, you are on Peak list price for Go. Same job Saturday afternoon is Off-Peak half price against the monthly dollar quota.

Go also uses rolling windows: 5 hours = 20% of the monthly quota, week = 50%, month = 100%. When the model quota is gone you can still use free models; Extra usage can fall back to Console balance if you enable it.

Beam Free, China notes, and “primarily international”

A few region facts that matter before you pick a path:

• Beam Free is documented as unavailable in countries including China (including Hong Kong and Macau). That blacklist sits on both Zen Free notes and the Go page.
• Go’s product copy says the service is designed primarily for international users and stable global access. That is not “China is blocked,” and it is also not “optimized for China.”
• Muse Spark Contributor lines follow Meta’s geographic policy and are a different story from DeepSeek Flash.

None of that rewrites the OpenCode DeepSeek V4.1 Flash price table. It only tells you which free neighbors on the same menu may 403 you.

What we actually ran (US egress — not China broadband)

On 2026-10-23 we installed OpenCode v2.1.3 on a Linux box whose public egress resolved to the United States (Fastly). Unsigned-in:

opencode run -m opencode/deepseek-v4.1-flash-free 'Reply with exactly: PONG'

It returned PONG in about two seconds. No Console top-up, no card, no OAuth.

This is not a China mainland broadband test. Do not quote it as “works on domestic ISP.” Domestic direct-connect behavior for Free / Beam / Muse is still on the “measure later” list.

BYOK DeepSeek API is a fourth bill

If you /connect → DeepSeek and paste a key from platform.deepseek.com, you pay DeepSeek’s own rates. That path does not inherit Zen Free, Zen $0.30/$1.20, or Go Peak/Off-Peak.

Useful when you want a stable China-oriented API bill. Confusing when you assume “I already paid Go so my BYOK key is free.” It is not.

Same idea for ChatGPT Plus/Pro or GitHub Copilot OAuth in OpenCode: usage counts against that subscription. Official providers docs treat ChatGPT / Copilot as first-class; Claude Pro/Max has an Anthropic ToS tension note — if you care about compliance, read Anthropic’s rules, not just the /connect menu.

How I would choose (without mixing prices)

• Want $0 first: use the timed Free ID. Expect limits and UI reset timers (v2.1.2+). Treat it as a trial lane.
• Already pay ChatGPT or Copilot: reuse that login in OpenCode before you dump $20 into Zen. Often cleaner.
• Want an auditable monthly $ quota per model: look at Go $10 / Plus $40 and the DeepSeek Flash $80 / $160 lines, then schedule work around Peak.
• Need something stable from China: prefer BYOK DeepSeek or local (Ollama / LM Studio / etc.), not a hope that every Free row is globally open.

If you are comparing coding agents themselves — open-source multi-provider vs Claude-locked vs OpenAI-locked — start with our Claude Code vs Codex guide. For “which subscription is worth it” in general, see which AI subscription is worth it. If you want DeepSeek offline instead of any cloud bill, run DeepSeek locally.

FAQ

Is OpenCode DeepSeek V4.1 Flash free forever?

No. The Free row is timed. The data page and Zen paid row show $0.30 / $1.20. Go can be half of that Off-Peak against a monthly dollar quota.

Does the data page mean Flash is free?

No. That page’s list price is the paid rate. Free is the separate Free-labeled entry.

When is Go Peak for this model?

UTC Mon–Fri 01:00–04:00 and 06:00–10:00. Beijing ≈ 09:00–12:00 and 14:00–18:00 on weekdays.

Can I use Beam Free in China?

Official Free notes list China (including HK and Macau) as unavailable for Beam Free. Separate from DeepSeek Flash pricing.

Does BYOK use OpenCode’s Zen price?

No. BYOK DeepSeek follows DeepSeek’s platform pricing.

More on this: Claude Code vs Codex · AI subscription picks · Run DeepSeek locally