Advanced AI models generate up to 50 times more CO₂ emissions than more common LLMs when answering the same questions
The more accurate we try to make AI models, the bigger their carbon footprint — with some prompts producing up to 50 times more carbon dioxide emissions than others, a new study has revealed.
Reasoning models, such as Anthropic's Claude, OpenAI's o3 and DeepSeek's R1, are specialized large language models (LLMs) that dedicate more time and computing power to produce more accurate responses than their predecessors.
Yet, aside from some impressive results, these models have been shown to face severe limitations in their ability to crack complex problems. Now, a team of researchers has highlighted another constraint on the models' performance — their exorbitant carbon footprint. They published their findings June 19 in the journal Frontiers in Communication.
"The environmental impact of questioning trained LLMs is strongly determined by their reasoning approach, with explicit reasoning processes significantly driving up energy consumption and carbon emissions," study first author Maximilian Dauner, a researcher at Hochschule München University of Applied Sciences in Germany, said in a statement. "We found that reasoning-enabled models produced up to 50 times more CO₂ emissions than concise response models."
To answer the prompts given to them, LLMs break up language into tokens — word chunks that are converted into a string of numbers before being fed into neural networks. These neural networks are tuned using training data that calculates the probabilities of certain patterns appearing. They then use these probabilities to generate responses.
Reasoning models further attempt to boost accuracy using a process known as "chain-of-thought." This is a technique that works by breaking down one complex problem into smaller, more digestible intermediary steps that follow a logical flow, mimicking how humans might arrive at the conclusion to the same problem.
Related: AI 'hallucinates' constantly, but there's a solution
However, these models have significantly higher energy demands than conventional LLMs, posing a potential economic bottleneck for companies and users wishing to deploy them. Yet, despite some research into the environmental impacts of growing AI adoption more generally, comparisons between the carbon footprints of different models remain relatively rare.
To examine the CO₂ emissions produced by different models, the scientists behind the new study asked 14 LLMs 1,000 questions across different topics. The different models had between 7 and 72 billion parameters.
The computations were performed using a Perun framework (which analyzes LLM performance and the energy it requires) on an NVIDIA A100 GPU. The team then converted energy usage into CO₂ by assuming each kilowatt-hour of energy produces 480 grams of CO₂.
Their results show that, on average, reasoning models generated 543.5 tokens per question compared to just 37.7 tokens for more concise models. These extra tokens — amounting to more computations — meant that the more accurate reasoning models produced more CO₂.
The most accurate model was the 72 billion parameter Cogito model, which answered 84.9% of the benchmark questions correctly. Cogito released three times the CO₂ emissions of similarly sized models made to generate answers more concisely.
"Currently, we see a clear accuracy-sustainability trade-off inherent in LLM technologies," said Dauner. "None of the models that kept emissions below 500 grams of CO₂ equivalent [total greenhouse gases released] achieved higher than 80% accuracy on answering the 1,000 questions correctly."
RELATED STORIES
—Replika AI chatbot is sexually harassing users, including minors, new study claims
—OpenAI's 'smartest' AI model was explicitly told to shut down — and it refused
—AI benchmarking platform is helping top companies rig their model performances, study claims
But the issues go beyond accuracy. Questions that needed longer reasoning times, like in algebra or philosophy, caused emissions to spike six times higher than straightforward look-up queries.
The researchers' calculations also show that the emissions depended on the models that were chosen. To answer 60,000 questions, DeepSeek's 70 billion parameter R1 model would produce the CO₂ emitted by a round-trip flight between New York and London. Alibaba Cloud's 72 billion parameter Qwen 2.5 model, however, would be able to answer these with similar accuracy rates for a third of the emissions.
The study's findings aren't definitive; emissions may vary depending on the hardware used and the energy grids used to supply their power, the researchers emphasized. But they should prompt AI users to think before they deploy the technology, the researchers noted.
"If users know the exact CO₂ cost of their AI-generated outputs, such as casually turning themselves into an action figure, they might be more selective and thoughtful about when and how they use these technologies," Dauner said.
Hashtags

Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
22 minutes ago
- Yahoo
Palo Alto to scoop up CyberArk for $25 billion to tackle AI-era threats
(Reuters) -Palo Alto Networks said on Wednesday it would buy Israeli peer CyberArk Software for about $25 billion, deepening its push to become a comprehensive provider of cybersecurity services amid rising AI-driven threats. The deal, Palo Alto' biggest yet and one of the largest tech takeovers this year, is the latest in a wave of consolidation as companies look to streamline vendors after facing breaches while relying on a patchwork of firms. It follows Alphabet's $32 billion acquisition of Israeli startup Wiz in March, and would broaden Palo Alto's cybersecurity offerings by adding identity security tools, bolstering its appeal to large enterprise customers. CyberArk investors will receive $45.00 in cash and 2.2005 shares of Palo Alto for each share they own under the deal, the companies said in a joint statement. The acquisition is expected to close in fiscal 2026 and will immediately add to Palo Alto's revenue growth, as well as gross margin. "The rise of AI and the explosion of machine identities have made it clear that the future of security must be built on the vision that every identity requires the right level of privilege controls," Palo Alto Networks CEO Nikesh Arora said. CyberArk specializes in privileged access management, technology that helps organizations safeguard sensitive systems by limiting and monitoring access to critical accounts. Its customers include Carnival Corp, Panasonic and Aflac. A surge in cyberattacks, including data breaches and ransomware, has driven demand for more comprehensive defenses. That has fueled interest in firms including CyberArk. The deal also comes as Palo Alto looks to accelerate its AI security push, with analysts at Scotiabank saying that it could use its large salesforce to drive adoption of CyberArk's tools -seen as critical for securing emerging agentic AI systems. Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
Yahoo
22 minutes ago
- Yahoo
SPQR TECHNOLOGIES LAUNCHES MACHINE REPUBLIC: FIRST AI SYSTEM WITH IMMUTABLE ETHICAL GUARDRAILS -- AS INDUSTRY FACES 'NO KILL SWITCH' CRISIS
As AI leaders admit 'No Kill Switch', SPQR deploys the world's first provable AI governance System. NEWARK, Del., July 30, 2025 /PRNewswire/ -- What if you couldn't trust your car's brakes, and every engineer said, "Just watch closer"? That's the moment the world now faces with artificial intelligence. Leading AI thinkers, from OpenAI, DeepMind, Meta and Anthropic to Geoffrey Hinton ("the Godfather of AI"), are sounding alarms about the industry's inability to guarantee control or transparency as AI systems grow in complexity and autonomy. In a recent technical paper, they claim that control is vanishing fast: "the world's last window into how AI makes decisions is vanishing fast… There is no guarantee that the current degree of visibility will persist." "There is no kill switch for AI. Tech fixes aren't enough. Physical controls are impossible. Patches won't hold," warns a recent industry analysis. Even Sam Altman, CEO of OpenAI, has publicly voiced fears about unpredictable risks from advanced AI models. (source) "There is no kill switch for AI. Tech fixes aren't enough. Physical controls are impossible. Patches won't hold. Emergency plans include "blowing up data centers" to halt a rogue AI. Are we already approaching Judgment Day? While global regulators from UNESCO to the EU scramble to define AI governance, SPQR Technologies Inc has proven it's possible, today. "Hope and narrative are not governance. The Machine Republic is not just a story, it's operational, provable, and designed for this moment," says Adam Massimo Mazzocchetti, Founder and Chief Architect of SPQR Technologies. The world's most powerful AI developers now admit they have no answers. SPQR Technologies offers a new path, one that is operational, verifiable, and governed by cryptographic constraints, rather than hopeful promises. The Machine Republic, developed by SPQR Technologies, is the first publicly evidenced AI governance system where ethical guardrails are not just written, they are imposed by quantum resistant, zero knowledge cryptography. These constraints are engineered into the core and cannot be bypassed by any developer, vendor, or third party. "Imagine brakes that cannot fail, not because you check them, but because they're engineered to stop the car, every time, no matter what," says Mazzocchetti "This isn't hand-waving about 'ethical AI.' It's enforcement, proven with empirical evidence and live operational video logs, at the code and legal level. That's what's missing in AI today, and that's what's running now." Why This Matters If the "AI window" closes, humanity loses not just trust, but control. The Machine Republic guarantees certain classes of misbehavior simply cannot happen, by design, not by hope. Governments, institutions, and the public no longer have to take promises on faith. They can verify, audit, and, if needed, intervene. SPQR's Proof In just five weeks, the Lex Machina Canon and related research have received hundreds of combined downloads and views across open public repositories including Zenodo, SSRN, and PhilPapers with three papers now ranked in the SSRN Top 10 for ORG: Regulation and PSN Rule of Law. Multiple papers are currently under active review, including a major paper at AI & Ethics (Springer Nature), where the editor has officially invited more than ten external reviewers to evaluate the manuscript (as confirmed in the publisher's submission system and [DOI link]). The system is operational, with public videos, logs, and evidence-backed research available now. Nine global provisional patents, including a wrapper patent, secure the intellectual foundation, ensuring real, enforceable governance, not just theory. SPQR's approach has already been recognized by leading legal theorist Lawrence Solum, who recently featured the work on the Legal Theory Blog. If the story so far proves anything, it's that this is not theory, this is a running system, helping shape humanity's future. SPQR Technologies has first reached out to policymakers, institutions, and state leaders worldwide out of respect for due process. With this release, the company now invites the public to participate. While the operational system is under controlled access, the underlying architecture and research are fully documented and available. The Machine Republic's research papers are open for examination, contribution, and challenge, from every corner of society. The Machine Republic is licensable, operational, and globally protected with foundational patents. This is a working system, ready to scale, partner, and deliver value. Real governance is bigger than business alone. SPQR Technologies invites experts, institutions, and forward looking partners to test, adapt, and help evolve this system for the world. While the code is currently proprietary, the company's roadmap is to open source. The ethical guardrails of The Machine Republic (Lex Suprema), "not a product, not a manifesto, but a scaffold for a republic that does not yet exist, but must", are already open for the world to help shape our collective future. This release is not just an announcement, but an open call for a future where governance is proven, not promised. SPQR Technologies: Not just promising responsible AI, delivering it, immutably at every level. To see the full journey and supporting documentation, visit: request a live demo by emailing media@ SPQR: Where AI governance is provable, not just promised. About SPQR Technologies: SPQR Technologies was founded by Adam Mazzocchetti to architect a new civic order for artificial intelligence, with the foresight of a founder, the teeth of an emperor, and the ethics of a philosopher. This is not just governance by design; it is governance by constitutional necessity. View original content to download multimedia: SOURCE SPQR Technologies
Yahoo
22 minutes ago
- Yahoo
Liberty Defense Selects K′(Prime) Technologies Inc. for Strategic Partnership and Increased Market Penetration in Canada
K′(Prime) Technologies Inc. to sell Liberty's next generation threat detection HEXWAVE™ system to the Canadian aviation and urban security market. WILMINGTON, Mass., July 30, 2025 (GLOBE NEWSWIRE) -- Liberty Defense Holdings Ltd. ('Liberty' or the 'Company') (TSXV: SCAN, OTCQB: LDDFF, FSE:E30), a leading technology provider of Artificial Intelligence (AI)-based next generation detection solutions for concealed weapons and threats, is pleased to announce it has entered into a strategic, non-exclusive distribution agreement with Calgary, Alberta-based K′(Prime) Technologies Inc. (K′(Prime)) to introduce HEXWAVE™ to the Canadian Aviation and Urban Security Market. K′(Prime) has over 20 years of experience in the sale, rental, and service of security equipment. As part of the distribution agreement, K′(Prime) will add the HEXWAVETM walkthrough people screening system, which uses AI and machine learning to seamlessly detect weapons and other threats in real-time, to its diverse security equipment and services portfolio. 'We are excited to partner with Liberty Defense to bring HEXWAVETM technology to the Canadian aviation and urban security market,' said Marc Tomlinson, K′(Prime)'s President. 'One of the primary reasons we partnered with Liberty Defense to sell the HEXWAVETM system is the company's shared commitment to innovation, meaningful results, and customer satisfaction; all of which will greatly benefit Aviation and Urban Security across Canada. We are deeply honored to provide best-in-class security technology. The HEXWAVETM system was selected as the perfect complement to our existing imaging and security services program, and we believe that incorporating cutting-edge walkthrough screening technology to detect non-metallic and metallic threats to security will help existing and new customers stay ahead of tomorrow's security risks.' 'We're thrilled to collaborate with K′(Prime), a respected leader in Canada's security sector,' said Bill Frain, CEO of Liberty Defense. 'It was clear after a visit to our Wilmington facility that we have a shared vision of advancing next-generation detection technologies while improving the overall security experience. We're proud that K′(Prime) selected the HEXWAVETM system to strengthen their aviation and urban security capabilities, especially in detecting emerging threats like non-metallic explosives and 3D-printed ghost guns. With K′(Prime)'s strong presence in aviation services across Canada, we look forward to expanding our market reach and driving meaningful revenue growth through this partnership.' For updates and news, please visit the Company website to subscribe to email alerts or follow Liberty Defense on social channels. On Behalf of Liberty DefenseBill FrainCEO & Director About Liberty DefenseLiberty Defense (TSXV: SCAN, OTCQB: LDDFF, FSE:E30) provides multi-technology security solutions for concealed weapons detection in high volume foot traffic areas and locations requiring enhanced security such as airports, stadiums, schools, and more. Liberty's HEXWAVE system uses millimeter wave, advanced 3D imaging, and AI technologies that have been secured under exclusive license and developed proprietarily to automatically detect all types of concealed metallic and non-metallic weapons and other prohibited items. Liberty is also developing the next generation of aviation passenger screening with the millimeter wave-based, High-Definition Advanced Imaging Technology (HD-AIT) body and shoe scanner technologies as part of its technology portfolio. Liberty is committed to protecting communities and preserving peace of mind through superior security detection solutions. Learn more: About K′(Prime) Technologies Inc.K′(Prime) Technologies Inc. (K′(Prime)) is a technology and scientific instrument services company, which brings value to the partners it represents by delivering quality services utilizing its vast network of technicians across Canada, enhancing sales, service coverage, and prioritizing customer satisfaction in all its activities. K′(Prime) has historically focused on laboratory, imaging, and commercial security instrumentation, supplies, and services. In 2022, K′(Prime) was awarded a five-year, $160 million contract with the Canadian Air Transport Security Authority (CATSA) to provide maintenance services for passenger screening (e.g., imaging) equipment at all Canadian airports. This expansion of the business has led to K′(Prime) delivering airport imaging equipment maintenance services for CATSA in every Canadian province and territory. This unique role positions K′(Prime) to support other businesses with similar maintenance and service requirements for imaging equipment and new technology. For further information about K′(Prime), please contact: Marc For further information about Liberty, please contact: Jay AdelaarSenior Vice President of Capital Markets, Liberty Defense604-809-2500jay@ For further sales information, please contact: Ian McNaughtonSenior Director Business development613-292-3669imcnaughton@ FORWARD-LOOKING STATEMENTSWhen used in this press release, the words 'estimate', 'project', 'belief', 'anticipate', 'intend', 'expect', 'plan', 'predict', 'may' or 'should' and the negative of these words or such variations thereon or comparable terminology are intended to identify forward-looking statements and information. Although Liberty believes, in light of the experience of their respective officers and directors, current conditions and expected future developments and other factors that have been considered appropriate, that the expectations reflected in the forward-looking statements and information in this press release are reasonable, undue reliance should not be placed on them because the parties can give no assurance that such statements will prove to be correct. Such statements and information reflect the current view of Liberty. There are risks and uncertainties that may cause actual results to differ materially from those contemplated in those forward-looking statements and information. By their nature, forward-looking statements involve known and unknown risks, uncertainties and other factors which may cause our actual results, performance or achievements, or other future events, to be materially different from any future results, performance or achievements expressed or implied by such forward-looking statements. There are a number of important factors that could cause Liberty's actual results to differ materially from those indicated or implied by forward-looking statements and information. Such factors include, among others: currency fluctuations; limited business history of the parties; disruptions or changes in the credit or security markets; results of operation activities and development of projects; project cost overruns or unanticipated costs and expenses; and general development, market and industry conditions. The parties undertake no obligation to comment on analyses, expectations or statements made by third parties in respect of their securities or their respective financial or operating results (as applicable). Liberty cautions that the foregoing list of material factors is not exhaustive. When relying on Liberty's forward-looking statements and information to make decisions, investors and others should carefully consider the foregoing factors and other uncertainties and potential events. Liberty has assumed that the material factors referred to in the previous paragraph will not cause such forward-looking statements and information to differ materially from actual results or events. However, the list of these factors is not exhaustive and is subject to change and there can be no assurance that such assumptions will reflect the actual outcome of such items or factors. The forward-looking information contained in this press release represents the expectations of Liberty as of the date of this press release and, accordingly, are subject to change after such date. Liberty does not undertake to update this information at any particular time except as required in accordance with applicable laws. Neither the TSX Venture Exchange nor its Regulation Services Provider (as that term is defined in the policies of the TSX Venture Exchange) accepts responsibility for the adequacy or accuracy of this press release.