Model

claude-2

2 entries · July 2023 – October 2023

Appears alongside

Featured in threads

Tracks

  • Benchmarks & progress 1
  • Models & capabilities 1

  1. SWE-bench paper published

    Built from 2,294 real GitHub issues across 12 Python repositories, the benchmark proved so hard that the best model of the day, Claude 2, solved under 2%.

    Benchmarks & progress

  2. Anthropic releases Claude 2

    A 100,000-token context window and a jump to 71.2% on the Codex HumanEval coding test, alongside a consumer web app opened to the US and UK.

    Models & capabilities