# OpenCode Go: A Tale of Three Bugs — What $10/Month Actually Gets You (and What It Doesn't) **Published: July 5, 2026 | Data-Driven Research Report** --- ## The Pitch vs. The Reality On paper, OpenCode Go sounds like a developer's dream: **$5 for your first month, $10/month thereafter** for reliable access to 13 of the best open-source coding models — including DeepSeek V4 Pro, Qwen3.7 Max, GLM-5.2, Kimi K2.7 Code, and more. Models hosted across US, EU, and Singapore data centers, all behind a single API key. But a deep dive into the official documentation and recent GitHub issue tracker reveals a more complex — and at times, deeply frustrating — picture. Over the past month, three distinct bugs have surfaced that collectively undermine the Go subscription's value proposition. Let's walk through each one. --- ## 📊 Part 1: What You're Actually Paying For First, let's get the numbers straight. The official docs at **opencode.ai/docs/go** lay out three rolling usage caps: | Cap Window | Dollar Limit | |---|---| | **5-hour rolling window** | **$12 of usage** | | **Weekly rolling window** | **$30 of usage** | | **Monthly rolling window** | **$60 of usage** | These are *dollar-value caps*, not request-count caps. So your mileage depends heavily on which model you use. ### The Model Economics Here's what $12 gets you in a 5-hour window: | Model | Requests per 5hr | Cost per Request (approx.) | |---|---|---| | **DeepSeek V4 Flash** | 31,650 | ~$0.00038 | | **MiMo-V2.5** | 30,100 | ~$0.00040 | | **DeepSeek V4 Pro** | 3,450 | ~$0.00348 | | **MiniMax M3** | 3,200 | ~$0.00375 | | **GLM-5.2** | 880 | ~$0.01364 | | **Qwen3.7 Max** | 950 | ~$0.01263 | The pricing per million tokens reveals the spread: - **Cheapest:** DeepSeek V4 Flash at **$0.14 input / $0.28 output** - **Most expensive:** Qwen3.7 Max at **$2.50 input / $7.50 output** - **Mid-range:** GLM-5.2 at **$1.40 input / $4.40 output** So a heavy GLM-5.2 user might burn through the 5-hour cap in under 900 requests, while a DeepSeek V4 Flash user could fire off 31,000+ in the same window. **The same $10 subscription delivers wildly different utility depending on model choice.** --- ## 🐛 Part 2: The Three Bugs That Changed Everything The official docs paint a tidy picture, but the GitHub issue tracker — specifically the **anomalyco/opencode** repository (183k stars, 3.9k issues) — tells the real story. Three issues in particular reveal systemic problems. --- ### Bug #1: The Silent Lapsing Trap (Issue #33451) > *"I subscribed to Go and made 3 payments totaling $25, but only the first payment ($5) activated my Go subscription. The subsequent payments ($10 + $10) were added to my Zen balance instead of renewing my Go subscription."* **Filed:** June 22, 2026 | **Status:** Closed **The timeline from the user's account:** 1. **April 21** — Subscribes to Go with $5 promo. Go activated. 2. **May 13** — Adds $10 to account *two days before* the renewal date, intending to fund the next month from balance. 3. **~May 21** — Go subscription expires. **No notification sent.** The API key *keeps working silently* — but now billing at Zen pay-as-you-go rates (which are higher than Go subscription rates). 4. **June 17** — Adds another $10 — again goes to Zen balance, not Go renewal. 5. **June 23** — Checks the dashboard. **No active membership shown.** Balance: $19.98 (and draining at premium rates). **The core mechanism failure:** OpenCode Go has **no way to renew a subscription from an existing Zen balance**. When you add funds, they flow to Zen by default. When Go lapses, there's no grace period, no notification — just a silent switch to metered billing. The user effectively paid $25 for $5 worth of subscription and $20 worth of overpriced pay-as-you-go credits they never asked for. > 💡 **The irony:** The official docs under "Usage beyond limits" state: *"If you also have credits on your Zen balance, you can enable the Use balance option in the console. When enabled, Go will fall back to your Zen balance after you've reached your usage limits."* This fallback is *documented as a feature* — but the user in this issue experienced it as a bug, with no opt-in, no notification, and no way to use Zen balance to *renew* the subscription. --- ### Bug #2: The Bad Gateway Blackout (Issue #35163) > *"HTTP 502 Bad Gateway — Cloudflare: 'The origin web server returned an invalid or complete response to Cloudflare.'"* **Filed:** July 3, 2026 | **Status:** Open On July 3, 2026, at approximately **12:50 UTC**, OpenCode Go's API endpoint (`https://opencode.ai/zen/go/v1`) began returning **HTTP 502 Bad Gateway** errors for all models. The Cloudflare layer was receiving incomplete responses from the origin server. **Key observations from the report:** - All models affected through the Go/Zen API endpoint - Each request retried 3 times, all exhausted - **Workaround worked immediately:** Switching to the *direct* DeepSeek API resolved the issue - Multiple duplicate reports flooded in within hours (#35148, #35156, #35160) - Related "Insufficient Balance" errors (#35149, #35159) suggested the root cause might be **upstream DeepSeek account/payment issues on the OpenCode Go backend** **The uncomfortable truth:** OpenCode Go is essentially a managed proxy to underlying providers (DeepSeek, MiniMax, Alibaba Cloud's Qwen, etc.). When the upstream billing breaks, all Go subscribers are locked out. The "reliable access" selling point takes a direct hit when you can bypass the issue entirely by using the provider directly. > 💡 **The lesson:** OpenCode Go adds convenience and consolidation — but it also adds a single point of failure. If you're relying on Go for production coding, you need a fallback provider configured. --- ### Bug #3: The Phantom Paywall (Issue #33495) > *"Users with an active OpenCode Zen balance are still being limited as if they are free-tier users. One account had a Zen balance of at least $20, but both accounts still hit the same free usage cap / 200 request limit."* **Filed:** June 23, 2026 | **Status:** Open This is perhaps the most disconcerting bug of the three. The user tested two accounts side-by-side: | Account | Zen Balance | Behavior | |---|---|---| | Account A | **$20+** | Hit FreeUsageLimitError at 200 requests | | Account B | **$0** | Hit FreeUsageLimitError at 200 requests | | **Result** | — | **Identical behavior** | The error message: ```json { "statusCode": 429, "error": { "type": "FreeUsageLimitError", "message": "Rate limit exceeded. Please try again later." } } ``` **What this means:** The free-tier usage limiter is being applied *before* the billing system checks whether the account has a Zen balance. The documentation says "pay-per-request after adding balance" — but in practice, the 200-request free cap gates everything, balance or not. This isn't just a minor annoyance. It means that users who *have paid for Zen credits* are being treated identically to users who haven't paid a cent. The "Use balance" fallback feature documented on the Go page **cannot function correctly** if the cap checker is broken at the infrastructure level. --- ## 🔍 Part 3: Connecting the Dots — A System Under Pressure When you look at these three bugs together, a pattern emerges: ### The Three Failure Modes of OpenCode Go | Issue | Failure Mode | Impact | Status | |---|---|---|---| | #33451 | **Billing routing** — payments misdirected to wrong balance, silent lapse | Financial: pay $25, get $5 value | Closed | | #33495 | **Cap enforcement** — Zen balance ignored, free tier limits applied to paying users | Functional: paid features don't work | Open | | #35163 | **Upstream dependency** — provider backend failure cascades to all Go users | Availability: complete service blackout | Open | **Three distinct layers of the stack — billing, rate-limiting, and upstream providers — all showing cracks within a 12-day window (June 22 – July 3, 2026).** ### The Subscription Math Doesn't Add Up (Yet) At $10/month with a $60 monthly cap, OpenCode Go is priced to compete with services like GitHub Copilot ($10-19/month) and Cursor Pro ($20/month). For a heavy DeepSeek V4 Flash user, the value is undeniable — 158,000 requests per month for $10. But the value proposition collapses if: - Your payments get trapped in the wrong balance bucket (Bug #1) - You can't actually use the credits you paid for because free-tier caps aren't lifted (Bug #3) - The entire service goes dark because an upstream provider's payment pipeline glitches (Bug #2) --- ## 🛠️ Part 4: Practical Recommendations ### For Current Go Subscribers 1. **Don't rely on the "Use balance" fallback.** Until Issue #33495 is resolved, having Zen credits does *not* guarantee you'll bypass the 200-request free cap. 2. **Always maintain a direct provider API key as backup.** The July 3 outage showed that switching to a direct DeepSeek API key is the fastest workaround when Go goes down. 3. **Monitor your subscription renewal manually.** Create a calendar reminder for your renewal date. Don't add funds expecting them to renew Go automatically — they'll go to Zen instead. 4. **Track your usage in the console.** The 5-hour cap of $12 is tight for expensive models. One intensive session with Qwen3.7 Max ($2.50/M input tokens) could burn through a significant portion of that window. ### For Prospective Subscribers - **If you plan to use cheaper models (DeepSeek V4 Flash, MiMo-V2.5):** Go is excellent value — you'll get tens of thousands of requests per month. - **If you plan to use premium models (GLM-5.2, Qwen3.7 Max):** The $10 subscription gets you roughly 880–950 requests per 5-hour window. Evaluate whether that meets your needs before committing. - **If you value reliability above all else:** Consider a direct provider subscription as your primary and use Go as a secondary/convenience option until the billing and cap infrastructure stabilizes. --- ## 📝 Conclusion OpenCode Go is an ambitious product — a curated, low-cost gateway to the best open-source coding models. The pricing is genuinely competitive, the model selection is excellent, and the vision of "no lock-in" is admirable. But the product is suffering from growing pains. Three distinct bugs in less than two weeks — two still open — suggest that the billing and rate-limiting infrastructure hasn't scaled to match the product's popularity. The **silent lapse** (Issue #33451) and the **phantom paywall** (Issue #33495) are trust-eroding issues that, if unresolved, could drive users toward direct provider relationships that bypass OpenCode entirely. For now, OpenCode Go is a **qualified recommendation**: great value for the right use case, but with enough rough edges that you should go in with eyes wide open — and a backup plan. --- *Data sources: [opencode.ai/docs/go](https://opencode.ai/docs/go), [GitHub Issue #33451](https://github.com/anomalyco/opencode/issues/33451), [GitHub Issue #35163](https://github.com/anomalyco/opencode/issues/35163), [GitHub Issue #33495](https://github.com/anomalyco/opencode/issues/33495). All data verified as of July 5, 2026.*