China's Zhipu AI has matched Claude Mythos performance on finding secu… — Fact Check: Inaccurate

What was claimed

China's Zhipu AI has matched Claude Mythos performance on finding security bugs at 1/4 the cost, representing a huge leap that resets the entire AI race against the US

Our verdict

Inaccurate

Sources state GLM-5.2 achieved results at approximately $0.17 per vulnerability found, which is roughly one-sixth (not one-quarter) the cost of comparable Claude-based workflows. One-sixth is significantly lower than one-quarter. Multiple posts and articles say Zhipu AI's new model is reportedly or comparable to Claude Mythos in security bug detection, but emphasize that independent benchmark details are limited and the scores cited are slightly below Mythos. Presenting this as a definitive match overstates what is currently only a reported, unverified comparison.

0 of 3 AI systems agree10 sources citedChecked Jun 29, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

Zhipu AI achieved this at one-quarter the cost of Claude Mythos (1/4 the cost).

Incorrect90%

1 of 2 AIs agree·ChatGPT: Can’t verify

This represents a huge leap that resets the entire AI race against the US.

Misleading77%

All 3 AIs agree

China's Zhipu AI has matched Claude Mythos performance on finding security bugs.

Misleading82%

2 of 4 AIs agree·ChatGPT: Verified, Perplexity: Can’t verify

Verified against 10 sources

Detailed Analysis

Parts of the statement are loosely grounded in recent reports, but they overstate certainty and impact. The cost and “resetting the entire AI race” claims are not supported by current evidence and appear exaggerated or speculative.

Why this verdict

Parts of the statement are loosely grounded in recent reports, but they overstate certainty and impact.
The cost and “resetting the entire AI race” claims are not supported by current evidence and appear exaggerated or speculative.

Claims checked

China's Zhipu AI has matched Claude Mythos performance on finding security bugs

Misleading78%

Multiple posts and articles say Zhipu AI's new model is reportedly or comparable to Claude Mythos in security bug detection, but emphasize that independent benchmark details are limited and the scores cited are slightly below Mythos.[1] [2] [5] [6] [3] Presenting this as a definitive match overstates what is currently only a reported, unverified comparison.

This claim may mislead readers without added context.

This represents a huge leap that resets the entire AI race against the US

Misleading80%

Sources frame the development as a significant milestone and evidence of intensifying global AI competition, but do not claim that it "resets the entire AI race" or fundamentally overturns US leadership.[1] [10] The statement is a rhetorical exaggeration, not a supported factual description of the broader AI landscape.

This claim may mislead readers without added context.

Zhipu AI achieves this performance at 1/4 the cost compared to Claude Mythos

Can’t verify55%

None of the available reports discussing Zhipu AI vs Claude Mythos mention a specific cost ratio like "1/4 the cost" or provide concrete pricing or token-rate comparisons.[1] [2] [3] [5] [6] Without reliable pricing data for both models and a clear benchmark context, this cost claim cannot be confirmed.

What other AI think

The claim contains both accurate and misleading elements. Zhipu AI's GLM-5.2 does match Claude on specific security bug detection tasks at lower cost, but the claim overstates the scope and certainty of this achievement. The comparison is limited to narrow benchmarks, not general performance, and the "resets the entire AI race" framing is hyperbolic.

Why this verdict

The claim contains both accurate and misleading elements.
Zhipu AI's GLM-5.2 does match Claude on specific security bug detection tasks at lower cost, but the claim overstates the scope and certainty of this achievement.
The comparison is limited to narrow benchmarks, not general performance, and the "resets the entire AI race" framing is hyperbolic.

Claims checked

At 1/4 the cost

Incorrect90%

This claim may mislead readers without added context.

Representing a huge leap that resets the entire AI race against the US

Misleading80%

While sources acknowledge this is a significant milestone in specialized domains, they note the model still trails on general benchmarks and that the comparison is limited to specific security tasks. The claim of 'resetting the entire AI race' is hyperbolic—sources describe it as intensifying competition in specialized areas, not fundamentally resetting the broader AI landscape.

This claim may mislead readers without added context.

China's Zhipu AI has matched Claude Mythos performance on finding security bugs

Misleading85%

Zhipu's GLM-5.2 matches Claude on specific security vulnerability detection tasks (IDOR detection: 39% F1 vs Claude Code's 32-37%), but this is a narrow, specialized benchmark. Sources emphasize the comparison is limited to specific cybersecurity tasks, not general Claude Mythos performance. The model still trails on broad general-purpose benchmarks.

This claim may mislead readers without added context.

The statement mixes a reportable technical claim with speculative and unverified assertions. Some reputable outlets and researchers have reported that Zhipu’s recent GLM-5.2 can match Mythos-level bug-finding on select benchmarks, but cost and geopolitical-consequence claims lack verifiable evidence. I rate the overall package as a mix of supported and unsupported claims.

Why this verdict

The statement mixes a reportable technical claim with speculative and unverified assertions.
Some reputable outlets and researchers have reported that Zhipu’s recent GLM-5.2 can match Mythos-level bug-finding on select benchmarks, but cost and geopolitical-consequence claims lack verifiable evidence.
I rate the overall package as a mix of supported and unsupported claims.

Claims checked

This represents a huge leap that resets the entire AI race against the US.

Misleading72%

While matching a capability on some benchmarks is notable, claiming it ‘resets the entire AI race’ is a broad geopolitical assertion that is subjective and unsupported by concrete evidence about long-term capability, deployment, or economic and strategic impact.

This claim may mislead readers without added context.

Zhipu AI achieved this at one-quarter the cost of Claude Mythos (1/4 the cost).

Can’t verify65%

I found no reliable public data or vendor-verified pricing/operational-cost comparison showing Zhipu’s bug-finding runs cost 25% of Anthropic’s; cost depends on many unpublished factors, so the precise 1/4 figure cannot be verified.

China's Zhipu AI has matched Claude Mythos performance on finding security bugs.

Verified78%

Multiple recent news reports and benchmark summaries state that Zhipu’s GLM-5.2 matched Mythos on certain security-vulnerability benchmarks or specific evaluations; this appears supported for select tests, not necessarily every metric or real-world deployment.

Share this result

Try DoubleChecks