@estebankiwiClaude Fable 5 has claimed the top spot on DeepSWE with a score of 70%. However, the performance gap between Fable 5 and GPT 5.5 is far more significant than just three percentage points. Fable 5 generates code that reads as if it were written by a senior engineer, while GPT 5.5 produces code that simply passes the tests. Both models deliver functional software, but only one delivers software that truly impresses.
查看原帖@estebankiwi 的照片· Jun 19, 2026· GPT
The image shows a horizontal bar chart comparing AI models. The focus is on a table with model names, performance metrics, and bars representing their performance. The model "claude-fable-5" is highlighted with a red rectangle around it and its corresponding bar is orange. The table columns are labeled "MODEL", "PASS@1", "AVG COST", "OUT TOK", and "STEPS". The model names listed are claude-fable-5, gpt-5.5, claude-opus-4.8, gpt-5.4, gemini-3.5-flash, and kimi-k2.7-code. No on-screen text stands out aside from the column headers and model names.
@estebankiwiClaude Fable 5 has claimed the top spot on DeepSWE with a score of 70%. However, the performance gap between Fable 5 and GPT 5.5 is far more significant than just three percentage points. Fable 5 generates code that reads as if it were written by a senior engineer, while GPT 5.5 produces code that simply passes the tests. Both models deliver functional software, but only one delivers software that truly impresses.
查看原帖@estebankiwi 的照片· Jun 19, 2026· GPT
The image shows a horizontal bar chart comparing AI models. The focus is on a table with model names, performance metrics, and bars representing their performance. The model "claude-fable-5" is highlighted with a red rectangle around it and its corresponding bar is orange. The table columns are labeled "MODEL", "PASS@1", "AVG COST", "OUT TOK", and "STEPS". The model names listed are claude-fable-5, gpt-5.5, claude-opus-4.8, gpt-5.4, gemini-3.5-flash, and kimi-k2.7-code. No on-screen text stands out aside from the column headers and model names.
