GPT-5.6's IQ Breaks 130 Genius Threshold for the First Time, Outsmarting 99% of Humans

marsbitPublished on 2026-07-16Last updated on 2026-07-16

Abstract

GPT-5.6 has reportedly achieved an IQ score of 136 on Tracking AI's proprietary offline test, surpassing the human "genius" threshold of 130 for the first time. This places it above an estimated 99% of humans in this specific metric. The test is designed to prevent memorization by using a private question bank. Multiple GPT-5.6 variants, including the vision model, consistently scored 136, leading competitors like Claude-5 Fable (130). User anecdotes suggest practical superiority over rivals in real-world coding and problem-solving tasks, such as building a physics simulation or a customer service app from a single prompt. While some speculate this approaches AGI for most users, the article notes IQ tests only measure a narrow slice of cognitive ability like pattern recognition. The significance lies in GPT-5.6's apparent ability to translate high test scores into effective task performance on novel, real-world problems.

Today, 99% of the global human population is actually outperformed by an AI in terms of IQ.

In Tracking AI's latest offline IQ test, multiple versions of the GPT-5.6 "full suite" soared to a score of 136.

This is the first time an LLM has pushed its IQ beyond the 130 mark.

In the distribution of human intelligence, 130 is the starting line for "genius," a level only about 1% of the global population can reach.

In other words, GPT-5.6 is smarter than 99% of humans.

GPT-5.6 Racks Up 136 Points, IQ Breaks "Genius Line" for the First Time

How credible is this "IQ"?

In fact, Tracking AI uses two sets of questions.

One is a public Mensa Norway-style test, available online for anyone to take, which models have already scored over 140 on.

The other is its own curated "offline question bank." It's not public, prevents leaks, and is specifically designed to block the loophole of "models memorizing answers in advance."

The 136 points GPT-5.6 achieved this time was on this most difficult, anti-cheating offline test.

On this offline leaderboard, the various variants of GPT-5.6 (including the vision version) collectively surged to 136 points, leaving all competitors far behind.

Close behind is Claude-5 Fable, with 130 points.

Further down, names like GPT-5.6 LUNA Max and Claude-4.8 Opus are still hovering between 117 and 123 points.

It's important to note that this 130-point threshold had never been crossed before.

Over the past year, wave after wave of models, from o3 to various flagship models, surged forward, all getting stuck at the 130-point door, with none truly stepping into the "genius range."

GPT-5.6 is the first to kick that door open.

And it didn't achieve this score alone; the entire SOL, TERRA family collectively soared to 136, with even the vision version keeping pace.

On Reddit, a developer conducted a hands-on test and concluded that GPT-5.6's intelligence feels significantly higher than GPT-5.5's.

In the following test questions, GPT-5.6 achieved outstanding results in the shortest possible time.

One test score might not be convincing enough, so what does GPT-5.6 look like when taken out of the exam room and put to real work?

More Than Just a Score: Putting GPT-5.6 to Work

Developer Amir Bohlooli fed the same physics simulation prompt to both Fable 5 and GPT-5.6 Sol, expecting to be crushed by Fable, but ended up being amazed by GPT.

It chose particle fluid simulation, with physics progressing in real-time rather than blindly running fixed calculations per frame, cramming CSS, interface, and rendering all into a single HTML file, and automatically hosting it as a shareable webpage. In short, a finished product.

Similarly, Ramanpal Singh used a single prompt to create a RAG-based customer service ticketing system.

Four roles, an admin backend, embeddable components, and it can automatically categorize complaints, recognize sentiment, and draft replies.

It built 5 such apps in one go, at a cost that was only a fraction of what Fable 5 would require.

The most vivid story is from Claire Vo.

A few days ago, she was stuck on a bug, thinking her own code was broken. After switching to GPT-5.6 Sol, she just threw out the line, "I just don't believe I can't fix this."

Sol fixed it in one attempt and even managed to get it running on other models.

Her assessment hit the nail on the head: Fable gets bogged down in technical absolute precision, becoming its own trap, while Sol's pragmatic approach gets the job done.

It has to be said, there's an entire real-world project between an AI that can solve test problems and an AI that can save the day.

Does This Count as AGI?

Some netizens have said, "For 99% of people, this is already AGI."

Looking at it calmly, this 136 score was achieved on a specific offline / Mensa Norway-style test by Tracking AI.

What it measures is mainly "standardized cognition" like abstract pattern recognition and logical reasoning.

The problem is: IQ tests were never designed for large models.

A Mensa exam paper can't measure a model's factual reliability, its tool-calling ability, or how dependable it is in real professional scenarios.

It only slices off one thin layer of "intelligence" and tells you how bright that slice is.

However, hands-on testing by users provides the other half of the answer: GPT-5.6 seems to be slowly merging the two capabilities of "solving test problems" and "getting things done."

The questions in standardized tests are ones models have likely seen thousands of times in their training data; the real test of skill is with those new problems they've never encountered and have no answers to copy from.

Whoever can hold steady there truly deserves the word "intelligence."

References:

https://x.com/davidpattersonx/status/2077049232490672458

https://trackingai.org/

This article is from the WeChat public account "新智元" (New AI Era), author: ASI Revelation

Trending Cryptos

Related Questions

QAccording to the article, what was the significant achievement of GPT-5.6 in the Tracking AI offline IQ test?

AGPT-5.6 achieved a score of 136 on the private, offline IQ test, which is the first time a large language model has crossed the 130-point 'genius' threshold.

QHow does the article describe the difference between the two sets of IQ tests used by Tracking AI?

ATracking AI uses two sets of tests: a publicly available Mensa Norway-style test that models have already scored highly on, and a private, offline question bank designed to prevent models from having seen the questions before, which is considered more difficult and cheat-proof.

QWhat practical examples are given in the article to demonstrate GPT-5.6's capabilities beyond test scores?

AThe article provides examples where GPT-5.6 successfully created a particle fluid simulation HTML file, built a RAG-based customer service ticket system with multiple features, and efficiently debugged a coding problem that other models failed to solve.

QWhat caution does the article mention about interpreting the IQ score of GPT-5.6?

AThe article cautions that the IQ test only measures a specific slice of intelligence, like abstract pattern recognition and logical reasoning, and does not assess a model's factual reliability, tool-use ability, or performance in real-world professional scenarios.

QWhat was a key distinction made between Claude-5 Fable and GPT-5.6 Sol in their approach to solving problems, according to developer feedback cited in the article?

AAccording to developer feedback, Claude-5 Fable was described as being overly focused on technical perfection, which could hinder practical problem-solving, while GPT-5.6 Sol was praised for its pragmatic approach that successfully got the job done.

Related Reads

Former CFTC Chairman, Circle President Tarbert: Preaching Long-Termism While Cashing Out $30 Million Himself

Former CFTC Chairman and Circle President Heath Tarbert has consistently advocated for a long-term vision in public, urging patience from investors as Circle’s stock price has fallen significantly from its peak. However, it has been revealed that since Circle’s IPO, Tarbert has continuously sold his CRCL shares through pre-arranged trading plans, cashing out approximately $30 million, without making any public market purchases. This contrast between his public messaging and personal actions has drawn criticism. Tarbert joined Circle in July 2023 as Chief Legal Officer, leveraging his regulatory experience to help guide the company through its IPO and expansion. Despite promoting stablecoins as long-term infrastructure, he established a 10b5-1 trading plan just before Circle went public, leading to substantial stock sales over the following year. In March 2026, he initiated another plan to sell more shares. His career trajectory highlights a pattern of moving between high-level regulatory roles and influential positions in the financial sector. After resigning as CFTC Chairman in early 2021, he joined Citadel Securities as Chief Legal Officer just 27 days later, during a period of intense regulatory scrutiny for the firm. He later joined Circle, aiding its efforts to navigate regulatory challenges for its public listing. While Tarbert's expertise in policy and compliance is valuable to companies like Circle, his actions—advocating long-term confidence while personally divesting—raise questions about the alignment between his public statements and his private financial decisions, leaving investors who followed his advice to bear the market risks.

marsbit13m ago

Former CFTC Chairman, Circle President Tarbert: Preaching Long-Termism While Cashing Out $30 Million Himself

marsbit13m ago

Gate Research Institute: The 'Wall Street-ization' Wave of Crypto Financial Products – Competition or Integration?

The article titled "Gate Research Institute: Are Crypto Financial Products Sparking a 'Wall Street' Wave—Competition or Convergence?" explores the evolving relationship between the crypto ecosystem and traditional finance (TradFi). The piece begins by reflecting on Bitcoin's original 2009 vision of decentralization, disintermediation, and moving away from banks. It then contrasts this with the 2024 landscape, where key crypto assets like Bitcoin are increasingly held through Wall Street products like ETFs issued by giants like BlackRock. The article questions whether this signifies that TradFi is systematically taking over the rights to issue, price, custody, and distribute crypto financial assets. The core argument is that this is not a zero-sum takeover but rather a bidirectional convergence where each side addresses the other's weaknesses. Crypto offers 24/7 global markets, programmable settlement, and open access but lacks compliant channels, institutional-grade custody, deep fiat liquidity, and mainstream distribution. TradFi possesses these but is constrained by legacy systems, limited operating hours, and slow settlement. Two primary convergence paths are highlighted: * **Path A (CEX to TradFi):** Exemplified by Gate, which has progressed from offering tokenized stocks and CFDs to providing direct, real stock trading (US, Hong Kong, South Korea) within its platform, using USDT. * **Path B (TradFi to Crypto):** Exemplified by Robinhood, which has integrated crypto trading, acquired exchanges like Bitstamp, and is moving traditional assets like stocks onto the blockchain via tokenization and its own Layer 2. Both paths are ultimately competing to become the next-generation, unified financial account—a "super account" where users can seamlessly trade cryptocurrencies, stocks, ETFs, RWA (Real World Assets), and tokenized treasury products in one interface. The growth of RWA and tokenized treasuries (e.g., BlackRock's BUIDL) is presented as the asset-layer fusion, providing stable, yield-bearing assets on-chain and acting as a bridge between the two worlds. In conclusion, the "Wall Street-ization" of crypto is framed as a mutual transformation. Decentralized ideals persist in the protocol layer, while at the application layer, a more efficient, global, and accessible unified capital market is emerging from this convergence. The future competition lies not between crypto exchanges and stockbrokers, but between platforms vying to offer the most comprehensive asset coverage, liquidity, and user experience within a single account.

marsbit17m ago

Gate Research Institute: The 'Wall Street-ization' Wave of Crypto Financial Products – Competition or Integration?

marsbit17m ago

Claude's Major New Feature: Screen Recording + Voice, Distilling Your Skills into AI Tasks in One Click

Claude has introduced a major new feature called "Record a Skill," available for Pro, Max, and Team users. This function, found in the Claude desktop app's CoWork menu, allows users to create reusable AI skills simply by recording their screen and providing voice narration while performing a task. Claude then automatically analyzes the recording and generates a functional Skill. A hands-on test confirmed the feature works seamlessly. Users start recording via the Skills manager, perform their workflow while verbally explaining the steps and logic, and avoid including sensitive information. After recording, Claude processes the content and creates the Skill, which can be saved and later invoked with a slash command (/). This eliminates the need for manual adjustments or writing complex instruction files. The innovation goes beyond mere efficiency. Previously, creating a Skill required writing a detailed SKILL.md file in Markdown—a significant barrier for non-technical users. "Record a Skill" bypasses this by directly capturing both actions and the implicit reasoning shared in the narration. This lowers the barrier to knowledge transfer and automation, addressing a core challenge in corporate knowledge management: the difficulty of getting experts to write and maintain documentation. However, the feature also highlights a shift in the nature of work. A case study from March 2026 showed a freelancer whose five-year client relationship was effectively replaced by a hand-coded Claude Skill automating their content workflow. With the even lower barrier of screen recording, the ability to distill personal expertise into automatable skills accelerates this trend. The "moat" for work is moving from simply knowing how to do a task to mastering tasks that are difficult or impossible to automate.

marsbit21m ago

Claude's Major New Feature: Screen Recording + Voice, Distilling Your Skills into AI Tasks in One Click

marsbit21m ago

Feeding AI "Noise" Can Also Boost Scores, This Work Enables Positive Transfer with Noise

Feeding "Noise" to AI Can Improve Performance: A Method Enables Positive Transfer from Noise This work, Semi-Supervised Noise Adaptation (SSNA), introduces a Noise Adaptation Framework (NAF) that challenges traditional transfer learning. Instead of requiring a labeled source domain of real data (e.g., images, text), NAF uses randomly generated Gaussian noise as the source. For a target task with C classes, it constructs C noise clusters by sampling from Gaussian distributions. Although this synthetic noise contains no semantic meaning, NAF trains it to form a discriminative class structure in a shared representation space—clustering same-class noise and separating different classes. The key is aligning this learned structure from the noise domain to the real, sparsely labeled target domain. A small number of target labels are still essential to establish the correspondence between noise clusters and actual classes. The training objective combines: 1) supervised loss on the few labeled target samples, 2) classification loss for the noise to build its structure, and 3) a distribution alignment loss (using Negative Domain Similarity) to minimize the gap between the noise and target domains in the shared space. Experiments show significant gains in few-label settings. With just 4 labels per class, NAF with a ResNet-18 backbone improves accuracy over a standard supervised baseline (ERM) by +12.35% on CIFAR-10, +7.61% on CIFAR-100, +4.38% on DTD-47, and +2.74% on Caltech-101. It also benefits fine-grained datasets and scales to ImageNet-1K (with 100 labels/class) and text classification (AG News). NAF can be integrated into existing semi-supervised methods like FixMatch for further gains. Ablation studies confirm the transferred benefit comes from the discriminative structure of the noise, not randomness itself. Collapsing all noise into a single point causes negative transfer, while increasing separation between noise cluster centers improves performance. The amount of noise per class is less critical once a basic structure forms. In conclusion, this work demonstrates that for positive transfer, the semantic content of source data may not be necessary. What can be effectively transferred is the *organizational structure* of categories within a representation space. This offers a promising alternative for scenarios where real source data is unavailable due to privacy, copyright, or procurement constraints.

marsbit23m ago

Feeding AI "Noise" Can Also Boost Scores, This Work Enables Positive Transfer with Noise

marsbit23m ago

Trading

Spot

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of S (S) are presented below.

活动图片