OpenCode Go Review: Plans, Limits, Command Code, Charm Hyper

OpenCode Go is a $10 a month subscription, with a $40 Go Plus tier, for curated open coding models in OpenCode and compatible agents. Command Code sells credit plans from $1 (Go) and $10 (GOAT), and Charm Hyper, the product some tools call HyperCharm, sells 100 free Hypercredits a month and a $20 plan. As of 4 October 2026 all three publish dollar-based limits, which makes them easier to compare than most plans.

What are OpenCode Go and OpenCode Zen?

OpenCode Go costs $10 a month and Go Plus costs $40. Both give access to the same list of about 30 models, including GLM, Kimi, MiniMax, Qwen, DeepSeek, MiMo, Grok and GPT 6 Luna, and only the limits differ. Limits are monthly dollar amounts per model, with 20% of the monthly limit available per five hours and 50% per week. Cheap models get large limits ($60 a month on Go for GLM-5.3-Flash and DeepSeek V4.1 Flash) and expensive ones small limits ($15 for Kimi K3 and DeepSeek V4 Pro).

OpenCode's request estimates make that concrete: Go allows 130,000 DeepSeek V4.1 Flash requests a month but 490 for Kimi K3. When you reach a limit the free models keep working, or you can enable "Use balance" to fall back to a Zen balance. Only one member per workspace can subscribe, and the service is designed for international users.

Zen is the pay-as-you-go side. You add a $20 balance, pay per request "with zero markups", and card fees pass through at cost (4.4% plus $0.30). It lists Claude, GPT, Gemini and open models at provider prices, for example Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens.

What is Command Code, and what is the GOAT plan?

Command Code is a coding agent CLI (npm i -g command-code) built for open models, with a learning feature it calls taste-1. Its pricing page lists Go at $1 a month with $10 of credits, GOAT at $10 with $70, Pro at $20 with $80, Max 10× at $100 with $150 and Max 20× at $200 with $300. Each adds a processing fee whose amount the page does not state. A separate Provider plan costs $15 and bills pay as you go through OpenAI and Anthropic compatible endpoints. Every plan except Go has API access.

GOAT exists, and its $70 needs a close reading. The GOAT docs say credits are per-model allowances: up to the full $70 on GPT-5.6 Sol, GLM-5.2, Tencent Hy3 and Qwen 3.8 27B, $60 on DeepSeek V4.1 Flash, but $10 on Claude Sonnet 5.5 and $20 on new models until better deals land. Rolling caps are $14 per five hours, $35 per week and $70 per month. The estimates are 154,000 DeepSeek V4.1 Flash requests a month against 746 for Sonnet 5.5, and Command Code says allowances can change at any time. Extra credits are bought at model cost and never expire. Most models are zero data retention by default, and CMD_ZDR=1 enforces it. For new subscribers, Go switched to per-model credits on 28 September, with caps of $3, $6 and $10.

What is HyperCharm?

We could not find an official product named HyperCharm. The official product is Charm Hyper at hyper.charm.land, made by Charm (Charmbracelet, Inc.), the company behind the Crush terminal agent. The name HyperCharm appears in third-party tooling: My Claude Code lists it as a provider with the key HYPERCHARM_API_KEY, and community extensions for the Pi agent use it too. Everything below describes Charm Hyper, from Hyper's own pages.

Hyper is inference for coding, in Charm's words "cost-effective, high performance, coding-optimized, open source models, with zero data retention". The free plan gives 100 Hypercredits a month, the $20 subscription gives "250 Hypercredits, refreshing daily", and prepaid bundles cost $5 for 100, $10 for 200 and $20 for 400, and never expire. One Hypercredit is currently 5¢, so the free plan is about $5 of tokens and a bundle returns exactly what you pay. The pages do not say how many daily refreshes a month holds, so we do not convert the $20 plan into a monthly figure. Hyper's public models endpoint listed 23 models on 4 October 2026 with per-million-token prices, for example DeepSeek V4 Flash at $0.20 input and $0.40 output, and Kimi K3 at $3.27 and $16.33.

Crush has first-class Hyper support. Claude Code connects through an Anthropic-compatible endpoint with ANTHROPIC_BASE_URL=https://hyper.charm.land, and Codex, OpenCode, Pi and Zed use the OpenAI-compatible endpoints at https://hyper.charm.land/v1. On privacy, the docs state no training on your data and a 30-day maximum retention policy, while the FAQ describes zero data retention agreements with cloud providers. Check which one applies to you.

How do they compare with a Claude Pro baseline?

Prices and wording are from each vendor's page on 4 October 2026, with Claude Pro as the baseline.

Plan	Price per month	Included usage, vendor wording	Models	Tools
Claude Pro	$20 ($17 annual)	"More usage" than Free, number not published	Opus, Sonnet, Haiku; Fable on credits	Claude Code, Claude apps
OpenCode Go	$10	$15 to $60 of list-price usage, by model	About 30, mostly open	OpenCode, validated agents
OpenCode Go Plus	$40	$60 to $240 by model	Same list	Same
Command Code Go	$1 plus fee	$6 to $10 by model	Open models, a few closed	Command Code CLI, no API
Command Code GOAT	$10 plus fee	$10 to $70 by model	63 models	CLI and API
Charm Hyper Free	$0	100 Hypercredits a month (about $5)	23 coding models	Crush, Claude Code, Codex, OpenCode, Pi, Zed
Charm Hyper	$20	250 Hypercredits, refreshing daily	Same	Same

Claude Pro publishes no allowance, so the baseline is price and tools, not tokens. On our usage-based analysis, $20 buys 1,430 Sonnet-class calls at API list prices, while OpenCode Go's $10 allows 130,000 DeepSeek V4.1 Flash requests. The plans cover different model classes. GOAT's Sonnet 5.5 allowance is 746 requests a month, fewer than a daily developer's 3,300 calls, so these plans do not replace frontier access if you depend on it.

Which tools work with them, and what do the terms say?

OpenCode says Go is designed for OpenCode and other coding agents that send similar requests, and it monitors traffic for abuse. Clients should identify with their own user agent and send a stable session ID. Validated clients include Claude Code, Codex, Pi, Hermes, ZCode, jcode and Kilo Code CLI, without a guarantee they keep working. Known problematic clients include GitHub Copilot Chat and Kimi Code. Command Code sells only through its own site ("Resellers? Nope.").

All of these sell open-model access on their own API keys. None gives you a Claude or ChatGPT subscription, so Anthropic's rules for subscription credentials are not in play. Zen and Command Code do sell Claude at API prices, which is an ordinary API route. Never share or resell an account.

Who suits which plan?

• OpenCode Go: developers who use OpenCode or want a cheap, predictable second agent on open models.
• Go Plus: heavy use of expensive open models. Kimi K3's limit is $60 against $15 on Go, four times higher.
• Command Code GOAT: people who like a CLI harness and dollar allowances per model, and who check which models are boosted.
• Command Code Go: a $1 trial with light use.
• Charm Hyper: Crush users, and teams that want master keys, sub-keys and per-user usage reports.
• Zen: teams that want admin roles and spend limits, or occasional use without a subscription.

What are the caveats?

• Allowances move. Kimi K3's boosted GOAT usage runs through 7 October, two OpenCode models are free for a limited time, and DeepSeek's zero-retention agreement is valid through 31 October 2026.
• The headline dollar figure is a best case. It applies to the best-discounted models, not to every model.
• Privacy differs by model. OpenCode's Muse Spark Contributor models train on your prompts and are limited to some regions.
• Open-model quality varies by task. Test on your own code; see our open-weight models article.

How does Danube Labs use these plans?

My Claude Code lists OpenCode Zen, OpenCode Go, Command Code and HyperCharm among its 57 providers, so one tier map can send hard work to a strong model and subagent work to a cheap one; see Claude Code with any model. It is independent and not affiliated with OpenCode. To compare plans for your own workload, read about My Claude Code or book a free intro call, and keep the full plan comparison open.