Fugu vs GLM-5.2 vs Mythos: Why AI benchmarks crown different winners

by Hansraj Agrawal | Jul 1, 2026 | Technology

The latest race among advanced artificial intelligence (AI) systems is increasingly being fought through benchmark scores, with Japan’s Sakana AI and China’s Z.ai (formerly Zhipu AI) reporting stronger performance than Anthropic’s Claude Mythos on select evaluations.

However, the comparison is not entirely straightforward because the systems being measured are designed for different purposes.

Different models, different strengths

Claude Mythos is Anthropic’s flagship frontier AI model, built as part of the company’s broader Claude family of products.

Anthropic offers different Claude models for different use cases, including Opus for complex reasoning and coding tasks, Sonnet for general-purpose workloads, and Haiku for faster and lower-cost applications. Mythos sits at the top end of that stack and is positioned as Anthropic’s most capable model for software engineering, reasoning, and agentic tasks.

On the other hand, Sakana AI’s Fugu is not a standalone frontier model in the traditional sense. Instead, it functions as an orchestration layer that coordinates multiple frontier AI models and dynamically routes tasks to whichever model is best suited to solve them. In its technical report, Sakana AI said Fugu models are trained to “adaptively and dynamically orchestrate a team of more powerful frontier agent workers” and can achieve performance beyond a single model through what it describes as collective intelligence.

China’s GLM-5.2 occupies yet another position in the market. Rather than focusing primarily on raw benchmark performance, Z.ai has marketed the model around long-horizon task completion, autonomous coding workflows and agentic execution.

Why the distinction matters

It matters because many of the benchmark results now being cited compare fundamentally different approaches to AI development. While Anthropic is evaluating the capabilities of a frontier foundation model, Sakana AI is testing an orchestration system built on top of multiple models, and Z.ai is focusing on agentic systems designed for longer-duration workflows.

As a result, the benchmark leaders vary depending on what is being measured.

Where the models differ

The benchmark comparisons being cited by Anthropic, Sakana AI and Z.ai are not entirely like-for-like because the systems being evaluated are built differently and, in some cases, the companies report results for different models within their product families.

Benchmark	What it measures	Claude Mythos / Opus	Fugu / Fugu Ultra	GLM-5.2
SWE-Bench Pro	Real-world software engineering	80.3% (Mythos)	73.70%	62.10%
Terminal Bench 2.1	Agentic coding via terminal use	80.4% (Mythos Preview)	82.10%	82.70%
GPQA Diamond	Graduate-level science reasoning	94.6 (Mythos Preview)	95.5	91.2
CharXiv Reasoning	Scientific charts and figures	86.1 (Mythos Preview)	86.6	Not reported
Humanity’s Last Exam*	Multidisciplinary reasoning	64.7% (with tools)	50.00%	54.7% (with tools)
FrontierSWE	Long-horizon coding tasks (20 hrs)	75.1% (Opus 4.8)	Not reported	74.40%
PostTrainBench	Long-horizon agentic tasks	37.2% (Opus 4.8)	Not reported	34.30%
MCP-Atlas	Multi-step agent workflows	77.8% (Opus 4.8)	Not reported	76.80%
Tool-Decathlon	Tool use and workflows	59.9% (Opus 4.8)	Not reported	48.20%
ExploitBench	Cybersecurity and vulnerability tasks	78.0% (Mythos)	Not reported	Not reported
*Scores for Humanity’s Last Exam are reported under different evaluation settings and may not be directly comparable. Anthropic and Z.ai report tool-enabled scores, while Sakana AI reports a text-only score.

What does the comparison show

Anthropic’s models continue to lead several software engineering, cybersecurity and agentic workflow evaluations. The company reported an 80.3 per cent score for Claude Mythos on SWE-Bench Pro, a benchmark that tests whether AI systems can resolve real-world software issues in code repositories. Anthropic’s Opus 4.8 also leads benchmarks such as FrontierSWE, MCP-Atlas and Tool-Decathlon, according to the figures disclosed by the respective companies.

Sakana AI reported stronger performance on some scientific reasoning and agentic coding benchmarks. According to the company’s technical report, Fugu Ultra scored 95.5 on GPQA Diamond, a benchmark that evaluates graduate-level scientific reasoning, compared with 94.6 for Mythos Preview. Fugu Ultra also achieved 86.6 on CharXiv Reasoning, ahead of Mythos Preview’s 86.1. On Terminal Bench 2.1, which evaluates how effectively models can interact with computing environments through terminal commands, Fugu Ultra scored 82.1 compared with 80.4 for Mythos Preview.

Z.ai’s GLM-5.2 also reported stronger performance than Mythos Preview on Terminal Bench 2.1, scoring 82.7. The company further reported competitive results on long-horizon coding and agentic workflow benchmarks, including 74.4 on FrontierSWE and 76.8 on MCP-Atlas.

However, Anthropic’s Opus 4.8 remained ahead on both benchmarks, posting scores of 75.1 and 77.8, respectively.

Why benchmark wins do not always translate into overall leadership

The results illustrate why multiple companies can simultaneously claim state-of-the-art performance.

A system that excels at software engineering may not necessarily lead in scientific reasoning, while a model optimised for long-duration autonomous execution may perform differently on knowledge or cybersecurity evaluations.

Anthropic’s models continue to hold advantages across software engineering, cybersecurity and several workflow-oriented benchmarks. Fugu has reported stronger performance on selected scientific reasoning and agentic coding tests, while GLM-5.2 has reported competitive or leading results on some agentic coding and long-horizon software engineering evaluations.

The results point to an increasingly fragmented AI scenario, where different systems are optimised for different tasks. Rather than producing a single dominant model, the latest generation of AI products is creating specialists in areas such as software engineering, scientific reasoning, cybersecurity and long-horizon agentic execution.

Source link

Anthropic launches 'Claude Science' AI workbench for scientific research

by Hansraj Agrawal | Jul 1, 2026 | Technology

Anthropic on Tuesday launched Claude Science, an AI research workbench, designed to help scientists streamline research, analyze data and manage complex computing workflows.

The workbench offers scientists a user interface specifically designed for conducting research.

The ‌launch is part of Anthropic’s life sciences and healthcare initiative, which the IPO-bound company has been developing since October 2025.

Here are a few details on the launch:

Claude Science combines databases, coding tools, compute and research workflows in one workspace, helping scientists analyze literature, run analyses, create figures and manuscripts, and trace results back to ‌their source code and environment.

The tool is pre-configured with more than 60 scientific databases and can render scientific artifacts such as 3D protein structures, genome browser tracks and chemistry ‌drawings, Anthropic said.

Claude Science runs on Anthropic’s existing ‌Claude models, which have undergone the company’s standard responsible ‌scaling and biosecurity evaluations.

Several research organizations and companies testing the platform in beta reported significant efficiency gains, Anthropic added.

Anthropic is also launching its own pre-clinical drug programs, focused on neglected diseases, the AI startup’s head of life sciences Eric Kauderer-Abrams ‌said during a press briefing.

“These are areas that are outside ‌the scope of what the traditional pharma and biotech landscape might consider attractive targets, but nonetheless have real burden associated with them,” Kauderer-Abrams said.

Source link

US Supreme Court to hear Apple's appeal of contempt in Epic Games lawsuit

by Hansraj Agrawal | Jul 1, 2026 | Technology

The US Supreme Court agreed on Tuesday to hear Apple’s bid to escape being found in contempt in its legal fight with “Fortnite” maker Epic Games after the iPhone maker was deemed in violation of a judicial order mandating sweeping changes to its lucrative App Store in the antitrust litigation.

The justices took up Apple’s appeal of a lower court’s ruling upholding ‌a decision by Oakland, California-based US District Judge Yvonne Gonzalez Rogers finding Apple in contempt in Epic’s 2020 lawsuit contesting App Store fees.

The Supreme Court is expected to hear the case in its next term, which begins in October.

The lawsuit by Cary, North Carolina-based Epic Games challenged Apple’s control over transactions in applications that use the company’s iOS operating system and its restrictions on how apps are distributed to consumers.

Apple and Epic have clashed for years over the rules governing Apple’s App Store.

The contempt ruling and the scope of Apple’s court-ordered obligations are the latest issues in the dispute to reach ‌the Supreme Court. Apple has said the legal issues in the litigation will affect how millions of app purchases are made.

Cupertino, California-based Apple mostly defeated Epic’s lawsuit, but was required in a 2021 injunction issued by Rogers to let developers include links in their apps directing users to non-Apple payment methods.

Apple allowed the links but adopted ‌new restrictions, including a 27% commission on developers for purchases made on payment systems outside ‌the App Store within seven days of clicking a link. Apple charges developers a 30% commission ‌for purchases within the App Store. Epic argued that the new 27% commission flouted the earlier injunction. In 2025, Rogers found Apple in civil contempt for violating the injunction.

The San Francisco-based 9th US Circuit Court of Appeals in December upheld the judge’s contempt finding but let Apple make new arguments about what commission it should be allowed to charge for digital goods bought in apps distributed through the App Store but paid for using third-party systems.

That ‌new effort has not yet begun in the district court in Oakland.

Apple has denied violating the judge’s ‌order and argued to the Supreme Court that the injunction should not be applied to millions of developers beyond Epic Games.

“Regulators around the world are watching this case to determine what commission rate Apple may charge on covered purchases in huge markets outside the United States,” Apple told the Supreme Court in a filing.

(Only the headline and picture of this report may have been reworked by the Business Standard staff; the rest of the content is auto-generated from a syndicated feed.)

Source link

South Korea accuses Google of abusing Android app market dominance

by Hansraj Agrawal | Jul 1, 2026 | Technology

South Korea’s antitrust regulator alleged on Wednesday that Alphabet’s Google abused its dominant position in the Android app marketplace to hinder competition and will recommend corrective measures and a financial penalty.

The Korea Fair Trade Commission’s (KFTC) Market Surveillance Bureau found Google’s alleged abuse of market dominance in the Android app marketplace affected 14.16 trillion won ($9.1 billion) ‌in revenue, the bureau said in a media briefing where it released its examiner’s report on the matter.

From July 2019 to March 2026, Google’s Games/Google Velocity Program, which it internally called “Project Hug”, offered domestic and overseas game developers financial support for using Google services such as Cloud, Ads and YouTube, provided that they launched games on Google’s app store on terms at least as favourable as rival app marketplaces, the report said.

The contracts were also structured ‌so that Google’s financial support increased progressively as developers generated more revenue through Google Play, creating stronger incentives to prioritise Google’s marketplace.

The programme significantly reduced developers’ incentives to distribute games through competing app stores, including South Korea’s OneStore, blocking rivals’ business activities and forcing ‌developers into de facto exclusive dealing with Google, according to the ‌report.

“Google Play competes fairly with other app stores and delivers numerous benefits ‌to developers and consumers in Korea. We have cooperated diligently with the KFTC’s investigation, and we will continue to show the Commissioners that there has been no violation of the law,” Google said in a statement to Reuters.

If the commission ultimately concludes that Google abused its market dominance, it may impose a fine of up to ‌6% of the relevant affected revenue of $9.1 billion.

Google has eight weeks from receiving ‌the examiner’s report to submit a written response and review the evidence. The bureau said it plans to convene the full commission and issue a final ruling promptly once Google’s due process rights have been fully observed.

(Only the headline and picture of this report may have been reworked by the Business Standard staff; the rest of the content is auto-generated from a syndicated feed.)

Source link

Meta loses bid to dismiss US states' lawsuit over child addiction claims

by Hansraj Agrawal | Jul 1, 2026 | Technology

A federal judge rejected Meta Platforms’ bid to dismiss a lawsuit by 29 US state attorneys general accusing it of designing Facebook and Instagram to addict children and knowingly concealing the harm from the public.

In a decision late Monday, US District Judge Yvonne Gonzalez Rogers in Oakland, California, denied ‌Meta’s motion to dismiss claims based on deception, unfair practices and violations of the federal Children’s Online Privacy Protection Act.

The judge also said Meta did not comply with that law’s notice and parental consent requirements, and granted summary judgment to the states on that issue.

Meta said in a statement: “We strongly disagree with these allegations and are confident the evidence will show our longstanding commitment to supporting young people.” In a separate statement, California Attorney General Rob Bonta called the decision a “critical win” in holding Meta accountable for fueling a mental health crisis among American children.

Gonzalez Rogers also oversees related multidistrict litigation by more than 2,600 individuals, school districts and local governments over whether social ‌media platforms such as Facebook, Instagram, Google and YouTube, Snapchat and TikTok addict children.

Meta downplays harms

The states said research has shown that children’s use of Facebook and Instagram could lead to depression, anxiety, insomnia, interference with education and daily life, and self-harm including suicide.

Meta countered that the attorneys general had no evidence it misled consumers about its platforms’ alleged addictiveness, including in ‌congressional testimony by Chief Executive Mark Zuckerberg.

The Menlo Park, California-based company said this was because “social media addiction” ‌is not an established psychiatric condition, and therefore statements that its platforms are not addictive could not be ‌false.

Meta also said it didn’t violate the children’s online privacy law because it directed Facebook and Instagram to a general audience, not just children under age 13.

Judge finds factual disputes about addictiveness

In a 38-page decision, Gonzalez Rogers found material factual disputes over whether Meta’s social media platforms are addictive, whether Meta falsely denied it designed them that way, and whether it “partially” directed the platforms at children.

“The AGs present a reasonable interpretation of [Meta’s] statements that Facebook and Instagram are not designed ‌in ways that cause teens to compulsively use the platforms to their detriment,” the judge wrote.

“To the extent plaintiffs’ evidence ‌shows that the platforms are in fact designed to do just that, a jury could reasonably find the statements were untrue to a reasonable person,” she added.

A trial over California, Colorado, Kentucky and New Jersey’s claims against Meta is scheduled for August 18, court records show.

(Only the headline and picture of this report may have been reworked by the Business Standard staff; the rest of the content is auto-generated from a syndicated feed.)

Source link

US govt lifts export restrictions on Anthropic's Fable 5 AI model

by Hansraj Agrawal | Jul 1, 2026 | Technology

By Shirin Ghaffary and Maggie Eastland

The US government removed foreign access restrictions on Anthropic PBC’s Fable 5 artificial intelligence model, clearing it for wider distribution after the startup resolved the Trump administration’s safety concerns.

The Commerce Department had imposed an export control rule via a private letter on June 12 requiring the company obtain US permission before allowing any foreign national, regardless of location, to access its powerful Mythos 5 AI model and Fable 5, a similar model intended for broader release. In response, the AI company disabled the models and has been in discussions with officials to satisfy their concerns.

Anthropic, in a post Tuesday on X, said it had received notice that Commerce was lifting the curbs on access to both models. Some of the restrictions on Mythos had been eased on June 26.

Anthropic said it would restore access to users beginning Wednesday.

“We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models,” the company said in its post.

In recent days, Anthropic’s executives including co-founder Tom Brown had been discussing the matter with Commerce Secretary Howard Lutnick and other officials to resolve the government’s immediate issues with its latest releases, Bloomberg previously reported.

The key to reversing the export controls was assuaging White House officials’ concerns about limiting the ability of bad actors to circumvent the models’ guardrails.

In a letter Tuesday to Anthropic viewed by Bloomberg, Lutnick said the company had pledged to “proactively deter and address security risks associated with the models.” The secretary said Anthropic had agreed to work with the government on “protocols and standards” for future models.

The maker of the Claude chatbot was valued at $965 billion after its latest funding round, making it one of the world’s most valuable private companies. Anthropic announced June 1 that it had filed confidentially for an initial public offering. The company is considering going to the public markets as soon as October, Bloomberg has reported.

Earlier this month, the company released Fable 5 as the first public-facing version of its Mythos-class model, but with guardrails aimed at containing the full range of its cyber capabilities. Fable 5 is prevented from responding to certain types of queries, including those related to cybersecurity and biology. In those cases, Anthropic has said its Claude chatbot will route responses through a different model, called Opus 4.8.

Though now reversed, the department’s export control directive marked the most significant intervention by the US government to date into an AI venture’s operations and sparked legal questions over whether export controls can be used to regulate AI model access. The decision to restrict the availability of Anthropic’s models also reverberated across the AI industry, particularly at rival startup OpenAI, which limited the release of a capable new AI model, GPT-5.6, under pressure from the Trump administration.

OpenAI Chief Executive Officer Sam Altman said Friday that at the request of the US government, the company would roll out its latest 5.6 Sol model in a preview version to select partners who were approved by the administration, before making it more available.

Tuesday’s decision to relax the restrictions on Anthropic’s Fable model also relieves tensions that were weighing on the startup’s already-tenuous relationship with the Trump administration. The company is suing the Pentagon after Secretary Pete Hegseth moved in March to designate it a supply-chain risk following a messy and unsuccessful contract renegotiation process.

The easing of the curbs on Fable 5 were reported earlier by Politico and Wired.

Source link

« Older Entries

Next Entries »

Times Top News

Times Top News