Google’s Gemini 4 Argon and Anthropic’s Claude Sonnet 5.5 debut at the same price point but cater to different workflows, with benchmarks highlighting their distinct advantages in speed, specialised tasks, and security applications.
Anthropic’s Claude Sonnet 5.5 and Google’s Gemini 4 Argon have entered the market at the same nominal price, but they are optimised for rather different work. Sonnet 5.5 arrived on 28 September as Anthropic’s mid-tier model, positioned for routine professional tasks, while Google unveiled Argon two days later as its new frontier model for coding, enterprise knowledge work and defensive cybersecurity. Google says Argon is being rolled out first to trusted cyber defenders through its Fairwind programme, with wider availability to follow, whereas Anthropic has already made Sonnet 5.5 available through its apps, API and major cloud platforms.
That overlap in pricing has made the first independent comparisons unusually important. Both models list at $2 per million input tokens and $10 per million output tokens, but the headline figures do not tell the whole story. Anthropic says Sonnet 5.5 is more than 30% faster than its predecessor and can lower task costs by up to 30% through fewer tool calls and lower token consumption. Google, by contrast, emphasises Argon’s 1 million token context limit and its suitability for long, multi-step workflows.
Independent benchmarks suggest the models are closely matched overall, but not in the same ways. On the Artificial Analysis Intelligence Index, Sonnet 5.5 leads with 56 points against 53 for Argon. On the Vals Index, which weights tasks by their economic importance in the United States, Argon moves ahead with 68.90% versus 67.04% for Sonnet 5.5. The result is a split verdict: Sonnet 5.5 looks stronger on the broader reasoning mix used by Artificial Analysis, while Argon appears better aligned with the kinds of finance, legal and automation tasks that Google highlighted at launch.
The clearest edge for Sonnet 5.5 is in agentic terminal coding. Both Artificial Analysis and Vals found that Sonnet outperformed Argon on Terminal-Bench 4, with Sonnet scoring 64% and 64.14% respectively, compared with Argon’s 57% and 57.58%. That pattern matters because terminal-based coding is one of the most demanding practical uses for current models. By contrast, the lighter code benchmarks were far closer, and in some cases Argon remained competitive rather than decisively behind.
Argon’s strongest results came in finance, legal and defensive security tests. It scored 65.40% on Vals’ Finance Agent v2 benchmark, ahead of Sonnet 5.5’s 58.10%, and it posted a striking lead on Harvey’s Legal Agent Benchmark, where it completed 19.58% of tasks compared with Sonnet 5.5’s 2.92%. On cybersecurity, Vals’ CyberBench v1.1 put Argon at 77.86% against Sonnet 5.5’s 59.58%. Google has also claimed that Argon can autonomously find, validate and patch vulnerabilities, although the company has limited access while it develops the model further.
The cybersecurity comparison is also shaped by product policy, not just raw capability. Anthropic routes some high-risk requests away from Sonnet 5.5 to its older Sonnet 5 model as a safety measure, which can affect benchmark results if fallbacks are counted as failures. Google is taking the opposite route by restricting Argon’s access to vetted defenders during its Fairwind phase. That means any security benchmark reflects both model quality and the company’s chosen safeguards, so the numbers should be read as a measure of deployed product behaviour rather than pure model power.
Cost per task is where Argon has looked more efficient in the published tests, despite the identical sticker price. The independent runs showed Argon completing benchmark tasks at a lower effective cost than Sonnet 5.5 when both were pushed into their higher reasoning settings, largely because Sonnet consumed far more output tokens on some tests. Even so, the comparison is not symmetrical: Anthropic’s model is already widely available, while Google has not said when Argon’s introductory pricing will end or when broader access will open. In practice, that makes Sonnet 5.5 the safer choice for teams that need a model now, while Argon looks more promising for organisations willing to wait for access and for a better fit on finance, legal and cyber-heavy workloads.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
