Benchmark Performance vs. Claude Opus 4.8

The claim that GLM-5.2 sits "within one percentage point" of Claude Opus 4.8 on agentic coding is partially supported — it depends heavily on which benchmark you look at.

• On FrontierSWE (a long-horizon coding benchmark), GLM-5.2 scores 74.4 vs. Claude Opus 4.8's 75.1 — a gap of less than one point .
• On Terminal-Bench 2.1, GLM-5.2 scores 81.0 vs. Opus 4.8's 85.0 — a four-point gap .
• However, on SWE-bench Pro, Claude Opus 4.8 leads by about 7 points (69.2% vs. 62.1%) .

The "within 1 point" narrative comes from a specific subset of agentic benchmarks and was reported by CNBC and other outlets . On the broader set of tests, Opus 4.8 still leads on the most difficult long-horizon software engineering tasks .

Pricing: A Fraction of U.S. Models

GLM-5.2's API is priced at $1.40 per million input tokens and $4.40 per million output tokens . Compared to Claude Opus 4.8 ($5/$25 per million tokens), the cost advantage is approximately 3.6x to 5.7x cheaper, not the full 6x sometimes claimed, but still dramatic . Compared to GPT-5.5, it is reported as roughly one-sixth the price .

For teams that want subscription access, the GLM Coding Plan offers tiers from roughly $10/month (Lite) to $80/month (Max), with seat-based pricing for teams .

MIT License: Confirmed and Commercial

GLM-5.2's open weights were published on Hugging Face under the MIT license, which allows unrestricted commercial use, modification, and self-hosting with no revenue caps or regional restrictions . Multiple sources confirm this is a genuine open-weight release, not a "research-only" license .

360 Security's Tulongfeng and Yitianzhen: China's Answer to Mythos

On June 24, 2026, at the ISC.AI 2026 conference in Beijing, Chinese cybersecurity firm 360 Security Technology unveiled two AI tools positioned as domestic alternatives to Anthropic's Mythos cybersecurity model :

• Tulongfeng — An AI system for autonomous software vulnerability discovery. Founder Zhou Hongyi explicitly called it "China's version of Mythos" .
• Yitianzhen — A system designed to automate cyber defense and incident response .

Important Caveat: The Capability Gap

Zhou acknowledged that Tulongfeng has a 20-30% capability gap compared to the original Anthropic Mythos model . He emphasized that the strategy is about building a professional attack-and-defense team rather than trying to match a single "genius hacker" approach . Tulongfeng reportedly identified over 3,400 vulnerabilities, with 105 confirmed by China's government .

This announcement came during a period when Anthropic's Mythos and Fable 5 models were suspended globally due to U.S. export controls, creating a vacuum that Chinese and other Asian AI firms moved to fill .

The U.S. Export Ban on Anthropic's Mythos and Fable 5: A Temporary Shift

On June 12, 2026, the Trump administration directed Anthropic to restrict access to its most advanced AI models — Mythos and its consumer variant Fable 5 — citing national security concerns . The directive was linked to suspicions that a China-linked group had accessed Mythos .

Anthropic responded by suspending global access to both models, arguing that the scope of the directive — which even applied to its own foreign-national employees — made partial compliance impossible .

This created an immediate market vacuum. TechCrunch reported on June 27 that "Asian AI startups launch Mythos-like models as Anthropic's export ban drags on," mentioning both 360's Tulongfeng and a Tokyo-based startup Sakana AI's Fugu model .

The Ban Was Lifted

As of July 1, 2026, the U.S. government lifted the export ban, and Anthropic began restoring access to Fable 5 and Mythos 5 . Commerce Secretary Howard Lutnick stated the company had agreed to "proactively detect and address security risks associated with the models" . A limited release of Mythos was also permitted on June 26 for select companies .

Claims That Could Not Be Verified

The following claims from the original question could not be independently verified from the available sources:

These claims are plausible within the competitive narrative but require independent confirmation before they can be treated as verified facts.

The Bottom Line

The verified evidence shows a genuine competitive shift in mid-2026:

• GLM-5.2 is the strongest open-weight coding model currently shipping, trading blows with Claude Opus 4.8 on specific agentic benchmarks at a fraction of the cost, under a permissive MIT license .
• 360 Security has developed explicit domestic alternatives to Anthropic's cybersecurity AI, though with an acknowledged 20-30% capability gap .
• The U.S. export ban on Mythos/Fable 5 created a temporary vacuum that Chinese competitors moved to fill, but the ban was lifted after roughly three weeks .

The broader narrative — about a rapidly closing aggregate benchmark gap, massive cost advantages, and widespread enterprise adoption — is supported in part but not in totality by the evidence available here. Some of the most striking claims remain unverified.

Benchmark Performance vs. Claude Opus 4.8

The claim that GLM-5.2 sits "within one percentage point" of Claude Opus 4.8 on agentic coding is partially supported — it depends heavily on which benchmark you look at.

• On FrontierSWE (a long-horizon coding benchmark), GLM-5.2 scores 74.4 vs. Claude Opus 4.8's 75.1 — a gap of less than one point .
• On Terminal-Bench 2.1, GLM-5.2 scores 81.0 vs. Opus 4.8's 85.0 — a four-point gap .
• However, on SWE-bench Pro, Claude Opus 4.8 leads by about 7 points (69.2% vs. 62.1%) .

The "within 1 point" narrative comes from a specific subset of agentic benchmarks and was reported by CNBC and other outlets . On the broader set of tests, Opus 4.8 still leads on the most difficult long-horizon software engineering tasks .

Pricing: A Fraction of U.S. Models

GLM-5.2's API is priced at $1.40 per million input tokens and $4.40 per million output tokens . Compared to Claude Opus 4.8 ($5/$25 per million tokens), the cost advantage is approximately 3.6x to 5.7x cheaper, not the full 6x sometimes claimed, but still dramatic . Compared to GPT-5.5, it is reported as roughly one-sixth the price .

For teams that want subscription access, the GLM Coding Plan offers tiers from roughly $10/month (Lite) to $80/month (Max), with seat-based pricing for teams .

MIT License: Confirmed and Commercial

GLM-5.2's open weights were published on Hugging Face under the MIT license, which allows unrestricted commercial use, modification, and self-hosting with no revenue caps or regional restrictions . Multiple sources confirm this is a genuine open-weight release, not a "research-only" license .

360 Security's Tulongfeng and Yitianzhen: China's Answer to Mythos

On June 24, 2026, at the ISC.AI 2026 conference in Beijing, Chinese cybersecurity firm 360 Security Technology unveiled two AI tools positioned as domestic alternatives to Anthropic's Mythos cybersecurity model :

• Tulongfeng — An AI system for autonomous software vulnerability discovery. Founder Zhou Hongyi explicitly called it "China's version of Mythos" .
• Yitianzhen — A system designed to automate cyber defense and incident response .

Important Caveat: The Capability Gap

Zhou acknowledged that Tulongfeng has a 20-30% capability gap compared to the original Anthropic Mythos model . He emphasized that the strategy is about building a professional attack-and-defense team rather than trying to match a single "genius hacker" approach . Tulongfeng reportedly identified over 3,400 vulnerabilities, with 105 confirmed by China's government .

This announcement came during a period when Anthropic's Mythos and Fable 5 models were suspended globally due to U.S. export controls, creating a vacuum that Chinese and other Asian AI firms moved to fill .

The U.S. Export Ban on Anthropic's Mythos and Fable 5: A Temporary Shift

On June 12, 2026, the Trump administration directed Anthropic to restrict access to its most advanced AI models — Mythos and its consumer variant Fable 5 — citing national security concerns . The directive was linked to suspicions that a China-linked group had accessed Mythos .

Anthropic responded by suspending global access to both models, arguing that the scope of the directive — which even applied to its own foreign-national employees — made partial compliance impossible .

This created an immediate market vacuum. TechCrunch reported on June 27 that "Asian AI startups launch Mythos-like models as Anthropic's export ban drags on," mentioning both 360's Tulongfeng and a Tokyo-based startup Sakana AI's Fugu model .

The Ban Was Lifted

As of July 1, 2026, the U.S. government lifted the export ban, and Anthropic began restoring access to Fable 5 and Mythos 5 . Commerce Secretary Howard Lutnick stated the company had agreed to "proactively detect and address security risks associated with the models" . A limited release of Mythos was also permitted on June 26 for select companies .

Claims That Could Not Be Verified

The following claims from the original question could not be independently verified from the available sources:

These claims are plausible within the competitive narrative but require independent confirmation before they can be treated as verified facts.

The Bottom Line

The verified evidence shows a genuine competitive shift in mid-2026:

• GLM-5.2 is the strongest open-weight coding model currently shipping, trading blows with Claude Opus 4.8 on specific agentic benchmarks at a fraction of the cost, under a permissive MIT license .
• 360 Security has developed explicit domestic alternatives to Anthropic's cybersecurity AI, though with an acknowledged 20-30% capability gap .
• The U.S. export ban on Mythos/Fable 5 created a temporary vacuum that Chinese competitors moved to fill, but the ban was lifted after roughly three weeks .

The broader narrative — about a rapidly closing aggregate benchmark gap, massive cost advantages, and widespread enterprise adoption — is supported in part but not in totality by the evidence available here. Some of the most striking claims remain unverified.