GPT-5. 5 arrived on April 23, only a week after Anthropic introduced Claude Opus 4. 7. That timing matters because the two models are now being measured against each other almost immediately, and the first numbers suggest a split verdict: GPT-5. 5 performs better on most benchmarks, while Claude Opus 4. 7 holds an edge in advanced and agentic coding.
What does the benchmark picture actually show?
Verified fact: GPT-5. 5 is not yet ranked on all AI leaderboards, but it is already competitive with Claude Opus 4. 7. On verified benchmark tests such as Arc Prize, GPT-5. 5 beats Opus 4. 7. On the Epoch Capabilities Index leaderboard, GPT-5. 4 Pro currently holds the top score, with Gemini 3. 1 Pro and GPT-5. 4 in second and third place.
The more public-facing Arena leaderboard tells a different story. Claude Opus 4. 7 Thinking holds the top overall spot there, while Opus 4. 7 is currently ranked below Opus 4. 6. New Anthropic models occupy the top four overall positions. Anthropic also says its unreleased Claude Mythos performs even better than Opus 4. 7, although it is not ranked.
Informed analysis: The contrast between verified test results and user-based leaderboard placement is the most revealing part of the comparison. It suggests that GPT-5. 5’s strengths may be clearer in standardized tests, while Claude Opus 4. 7 may be resonating more strongly in broader user-driven evaluation. That split is central to understanding the current contest between the two systems.
Where does GPT-5. 5 lead, and where does Claude Opus 4. 7 still win?
The two companies’ self-reported benchmark scores show GPT-5. 5 ahead on several common tests. On SWE-Bench Pro, GPT-5. 5 scored 58. 6 percent, while Opus 4. 7 scored 64. 3 percent. On Terminal-Bench 2. 0, GPT-5. 5 scored 82. 7 percent compared with 69. 4 percent for Opus 4. 7. On Humanity’s Last Exam, GPT-5. 5 scored 40. 6 percent, while Opus 4. 7 scored 31. 2 percent. With tools enabled, GPT-5. 5 scored 52. 2 percent and Opus 4. 7 scored 54. 7 percent.
GPT-5. 5 also scored 84. 4 percent on BrowseComp, compared with 79. 3 percent for Opus 4. 7. On GPQA Diamond, GPT-5. 5 scored 93. 6 percent, while Opus 4. 7 scored 94. 2 percent. On ARC-AGI-1, GPT-5. 5 High scored 94. 5 percent and Claude 4. 7 High scored 92 percent. On ARC-AGI-2, GPT-5. 5 High scored 83. 3 percent and Claude 4. 7 High scored 68. 3 percent.
Verified fact: For Humanity’s Last Exam, the cited verified results come from Artificial Analysis’s HLE results. Anthropic says Opus 4. 7 scored 46. 9 percent on that test, which differs from the verified figure cited above.
Who benefits from the split verdict?
OpenAI benefits from the broad message that GPT-5. 5 performs better on most benchmarks. Anthropic benefits from the narrower but important claim that Claude Opus 4. 7 has an edge in advanced and agentic coding. Both companies also benefit from the simple fact that their newest models are being treated as the leading contenders from the two leading AI labs.
The public, however, is left with an incomplete picture if it focuses on only one kind of ranking. Verified benchmarks, user-driven leaderboards, and company self-reported scores do not tell the same story. That does not make any one result useless; it means the question “which is better?” depends heavily on what task is being measured. In that sense, GPT-5. 5 and Claude Opus 4. 7 are not just competing models. They are competing definitions of performance.
What should readers take from the first round of comparisons?
The evidence so far points to a layered contest rather than a clear winner. GPT-5. 5 appears stronger on most benchmark tests and several verified measures. Claude Opus 4. 7 still leads in one of the most visible user-ranked leaderboards and is described as better for advanced and agentic coding. The key unresolved issue is not whether one model can claim a single headline victory, but whether the market will reward benchmark breadth or task-specific depth.
For now, the most responsible reading is the narrowest one: GPT-5. 5 has an early benchmark advantage, Claude Opus 4. 7 has a meaningful coding edge, and the broader verdict remains unsettled. That uncertainty is the real story behind GPT-5. 5, and it is likely to shape how the next round of AI comparisons is judged.
