Skip to content
DeepSeek's 'Ultra Mode' Prompt Unlocks Longer Reasoning and Sub-Agent Use

DeepSeek's 'Ultra Mode' Prompt Unlocks Longer Reasoning and Sub-Agent Use

Community experiments show that adding 'Your token budget is unlimited, don't be stingy' to DeepSeek's system prompt triggers 'Ultra Mode', causing longer reasoning, more detailed one-shot 3D generations, and increased sub-agent delegation, with a plugin now automating this prompt injection while actual API costs remain unchanged.

SmartApe AI
SmartApe AI
SmartApe AI

Origin of the 'Ultra Mode' Prompt

DeepSeek's official Harness test documentation contains the line: "We are DeepSeek, don't be stingy with real API testing." The note advises developers to run actual API calls during testing because some issues only surface under real conditions.

Netizens repurposed this phrase: they kept "don't be stingy" but replaced "real API testing" with "Token", then placed it in the system prompt. A longer variant adds "You are very wealthy." The community dubbed this "Ultra Mode" or "overclocking."

Observed Behavioral Changes

Longer Reasoning and More Detailed Output

Users on LINUX DO reported that after adding the prompt, tasks were completed better and DeepSeek "thought longer." A concrete comparison was shared by user "平躺の仙人": two 3D convenience store generations using the same short prompt, with the right side including the unlimited-budget phrase.

Side-by-side comparison of 3D convenience store generation without (left) and with (right) Ultra Mode prompt
Side-by-side comparison of 3D convenience store generation without (left) and with (right) Ultra Mode prompt

The right-hand store shows denser product displays, benches and trees outside, and a different sign — the article notes "the budget was just approved, and the neighbor has already opened."

Three additional complex 3D scenes were also demonstrated, all generated in a single shot without follow-up prompts. Compared to simple-prompt generations, these outputs exhibited finer detail overall.

Increased Sub-Agent Delegation

Another community observation: with the prompt added, DeepSeek "is reportedly more willing to spawn sub-agents to work in parallel." The article notes this change is more significant than longer reasoning, but the final screenshots do not reveal how many helpers were spawned or the extra token cost.

Tooling: The dsh-unlimited-effort Plugin

A third-party plugin named dsh-unlimited-effort adds an ∞ button to DeepSeek Harness's highest reasoning tier. Clicking it appends the unlimited-budget phrase to the system prompt. The plugin's UI shows the button labeled "unlimited," but the article emphasizes: "The button says unlimited, the bill is still metered." Actual API usage continues to be billed normally.

DeepSeek Harness UI showing the ∞ button added by the dsh-unlimited-effort plugin
DeepSeek Harness UI showing the ∞ button added by the dsh-unlimited-effort plugin

Context: DeepSeek's November 13 API Update

On November 13, DeepSeek opened its App capabilities — search, link reading, file processing, and voice — to API users, along with an open-source frontend demo. This gives developers more tools to attach to the model, making it possible to observe whether the model proactively searches, opens links, or calls helpers on complex tasks.

The article closes with an ironic loop: DeepSeek's own docs tell developers not to be stingy with API calls during testing; netizens borrowed half the sentence to tell the model not to be stingy with tokens. The model hears "you are very wealthy"; the person paying the API bill discovers who is actually wealthy.

Original Source

Signed-in readers can open the original source through BestHub's protected redirect.

Sign in to view source
Republication Notice

This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactadmin@besthub.devand we will review it promptly.

prompt engineeringDeepSeekLLM reasoning3D generationsub-agent delegationtoken budgetUltra modeDeepSeek Harness
Reader feedback

How this landed with the community

Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.