gpt-6IMAGE
New RealSWE benchmark puts GLM-5.3 within reach of Astra
On the new RealSWE benchmark, Fable 5.1 scores 38.8%, GPT-6 Astra 33.8% and GLM-5.3 28.8%, showing open models closing in on the frontier.
Zain
@zainhasArchived 2026.09.14Originally posted 2026.09.12
Zain @zainhasOriginally posted 2026.09.12
I did not expect GLM 5.3 to be in spitting distance of Astra/Fable5.1 New RealSWE benchmark: > Fable 5.1: 38.8% > GPT-6 Astra: 33.8% > GLM-5.3: 28.8%
x.com/zainhas/status/2098570172600578072Archived 2026.09.14 15:53