Editorial illustration for Claude Mythos USD 36,428 for 122 exploit episodes; GPT‑5.5 USD 3,075 for 123
Claude Mythos USD 36,428 for 122 exploit episodes;...
Anthropic's latest AI model costs twelve times more than OpenAI's to do roughly the same job. The question isn't which is better, but whether the price difference even matters in the long run.
Claude Mythos ran 122 browser exploit episodes for $36,428. GPT-5.5 via Codex did 123 for $3,075. The UK's AI Safety Institute says Mythos performs somewhat better. That's the polite, expensive way of saying you get a slightly nicer result for a lot more money.
The price tags differ sharply: the full Mythos test run across 122 episodes cost about $36,428, according to ExploitBench. GPT-5.5 via Codex ran 123 episodes for roughly $3,075, about twelve times cheaper.
The raw numbers invite a simple conclusion. OpenAI can just buy more compute for GPT-5.5 and catch up. This is a contest that could be settled with a bigger check.
But the transcripts from Mythos complicate that. Seunghyun Lee reviewed them. He saw the model working like a competent, if expensive, security researcher.
It built an exploit humans had written off as too complex. That's the quiet part of the benchmark. Not the cost, but the capability.
Both models are now autonomously developing real weaponry. The price of entry is falling fast. The barrier between a clever proof-of-concept and a deployed tool is already gone.
Related Reading
Browse all Anthropic news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic Trump cracks down on Anthropic after Amazon tip; staff largely foreign
Open Source
Related Topic Claude Mythos highlights EU AI safety gaps, says researcher Caroli
Policy & Regulation
From Archives Use Automated Dashboards and Weekly Review Cadence for GenAI Interviews
From Archives OpenAI partners with Malta to offer ChatGPT Plus to every citizen
From Archives Tool Highlights Time‑Consuming Stalls and Faulty Calls in Claude Code
From Archives Claude targets agent control plane while Microsoft stays enterprise default
Common Questions Answered
What is the cost difference between Claude Mythos and GPT-5.5 for running browser exploit episodes?
Claude Mythos cost $36,428 to run 122 browser exploit episodes, while GPT-5.5 via Codex cost only $3,075 to run 123 episodes. This means Anthropic's Claude Mythos costs approximately twelve times more than OpenAI's GPT-5.5 to perform roughly the same task.
According to the UK's AI Safety Institute, how does Claude Mythos compare to GPT-5.5 in performance?
The UK's AI Safety Institute determined that Claude Mythos performs somewhat better than GPT-5.5, though the performance advantage is modest relative to the significant price difference. This means users get slightly nicer results with Mythos but at a substantially higher cost.
What specific capability did Claude Mythos demonstrate that impressed reviewer Seunghyun Lee?
Claude Mythos successfully built an exploit that humans had previously written off as too complex to develop. This demonstrated that the model could work like a competent security researcher, performing autonomous development tasks that went beyond what was initially considered feasible.
Why might OpenAI's GPT-5.5 potentially catch up to Claude Mythos despite the current performance gap?
According to the article, OpenAI could potentially close the performance gap by simply purchasing more compute resources for GPT-5.5. This suggests that the capability difference between the models could be resolved through increased computational investment rather than fundamental architectural differences.
Further Reading
- Papers with Code - Latest NLP Research — Papers with Code
- Hugging Face Daily Papers — Hugging Face
- ArXiv CS.CL (Computation and Language) — ArXiv
