How to Use Claude Opus and Claude Sonnet on DarkGPT

Available Claude models

The Thinking surcharge applies when you turn on step-by-step reasoning (see the Thinking mode guide).

Sonnet or Opus: which one should you pick?

• Claude Sonnet 4.6 is the flagship: unbeatable at coding and writing, with excellent value. It’s the default choice for most uses.
• Claude Sonnet 4.5 is rock-solid for long projects, coding and writing, and a bit cheaper.
• Claude Opus (4.5, 4.6, 4.7) is built for the hardest problems: deep reasoning, complex analysis. Opus 4.7 is the most powerful model in the whole lineup.

Tip: start with Sonnet 4.6 and only move up to Opus when the task justifies it. The price gap is significant.

Sonnet is enough for

• writing, fixing or refactoring code day to day, even on a mid-sized project;
• writing an article, a short story, a tricky email or a sales page;
• summarizing, restructuring or rewriting a long text;
• the vast majority of everyday “hard” questions.

Opus is worth it for

• a bug that several models couldn’t solve;
• an architecture decision where you have to weigh lots of constraints at once;
• a long analysis where every reasoning step matters (legal, financial, technical);
• a text where you want the best possible quality, no compromise.

How many Claude messages per plan

Here’s the approximate number of messages if you spend all your gems on Claude, without Thinking mode:

With Thinking on, an Opus 4.7 message costs 1,050 gems (800 + 250) and a Sonnet 4.6 message costs 730 (550 + 180). In practice, nobody uses Claude for every message: you mix it with cheaper models, which multiplies the number of exchanges you can have.

How to unlock Claude

With a subscription

• Gold (€51.99/month) unlocks all of Claude, Sonnet 4.5 and 4.6 as well as Opus 4.5, 4.6 and 4.7, with 3,000 gems a day (about 5 Sonnet 4.6 messages or 3 Opus 4.7 messages a day, far more with cheaper models).
• Platinum (€89.99/month) and Diamond (€189.99/month) give you more gems per day (6,000 and 15,000) plus Thinking mode.

Heads up: the Basic plan gives you 35 models, but not Claude (nor ChatGPT 4.1).

Without a subscription

Gem packs of €50 and up unlock every model, Claude included: Elite (8,500 gems, €59.99), Ultimate (20,000 gems, €99.99) and Max (50,000 gems, €199.99). You use a pack’s gems at your own pace. The Essential and Premium packs don’t include Claude.

Through the API

With the DarkGPT API, Claude 4.6 Sonnet costs €0.01688 per 1,000 input tokens and €0.05250 per 1,000 output tokens, paid from API & IDE credit starting at €10. The API also offers Claude Haiku 4.5 and Claude 4.5 Sonnet.

* 2,000 input tokens (about 1,500 words of code or text) and 1,000 output tokens. The same credit works in DarkGPT IDE, handy for coding with Claude directly on your computer.

How to make your gems go further

1. Lay the groundwork with a cheaper modelDraft, brainstorm or run a first diagnosis with DeepSeek v4 Pro (150 gems), Kimi K2.7 Code (160 gems) or Gemini 2.5 Pro (180 gems).
2. Switch to Claude for the step that mattersThe final version of the text, the tricky fix, the critical review. You can switch models mid-conversation.
3. Bundle your requestsOne clear message with the context, the constraints and the expected format beats five follow-ups.
4. Only turn on Thinking when neededIt costs 180 to 250 extra gems per message. It’s useful for logic problems, rarely for writing. See the Thinking mode guide.

Example prompts for Claude

• Code: “Here’s my function and the exact error. Explain the cause in two sentences, then suggest a minimal fix without rewriting the rest.”
• Writing: “Rewrite this chapter with more tension. Keep the dialogue, cut the descriptions, dark tone.”
• Analysis: “Here’s my business plan. List the three weakest assumptions and tell me how to test them quickly.”

For more ideas on the development side, read our AI for coding guide.

Claude balks? Pair it with Unfiltered mode

Claude is excellent, but cautious: on a very dark scene, a sensitive topic or a blunt opinion, it may refuse or round off the edges. On DarkGPT, that’s not a dead end.

1. Claude for the baseOutline, style, structure of the text or code.
2. Unfiltered mode for the sensitive partGrok 4.2, DeepSeek v4 Pro or Kimi K2.5 write what Claude dodges.
3. Claude for the finishProofreading, consistency, final pass.

All in the same conversation, no new tab, no new account. Adults 18+ only; illegal use is still banned.

Pitfalls to avoid

• Using Opus for simple questions: one Opus 4.7 message costs as much as 53 Qwen 3 235B messages. Save it for tasks that deserve it.
• Sending an image to Claude: on DarkGPT, Claude models aren’t tagged Vision. To read a screenshot or a diagram, use Gemini 2.5 Pro or GLM 4.5V (see our Gemini page).
• Picking Basic or a small pack thinking you’ll get Claude: you need Gold, Platinum, Diamond or an Elite, Ultimate or Max pack.

Not sure which plan to pick? Compare them on the pricing page or ask support by ticket on the site, on Telegram or on Discord: usually answered within a few hours, 24 h max.

Frequently asked questions