Dispatch
Banned Claude Fable 5 Still Confirmed #1 on DeepSWE and SWE-Bench Pro While Unbuyable Worldwide
Despite remaining suspended under the June 12 US export-control directive, Claude Fable 5 was confirmed #1 on DeepSWE (70% PASS@1, +3 over GPT-5.5) and SWE-Bench Pro (80.3%) by Datacurve on June 19, and scored 83.1% on Terminal-Bench 2.1 (June 17). For builders: the top-scoring coding model on multiple SWE benchmarks is currently uncallable, which is why GLM-5.2 and GPT-5.5 are absorbing displaced demand — plan agent capacity around models you can actually hit via API today.
↳ Follow the thread