# GPT-6 Astra vs Qwen3 235B A22b Instruct 2507

As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in 5 of 5 shared categories: General LLM, Reasoning, Long context, Factuality, and Abstraction (ARC).

GPT-6 Astra: $20 per 1M blended tokens.

Qwen3 235B A22b Instruct 2507: price per 1M tokens is unset in the catalog.

GPT-6 Astra: closed weights.

Qwen3 235B A22b Instruct 2507: open weights.

- Which to pick, General LLM: As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in General LLM at #25 against #90.
- Which to pick, Reasoning: As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in Reasoning at #23 against #80.
- Which to pick, Long context: As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in Long context at #22 against #88.
- Which to pick, Factuality: As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in Factuality at #25 against #82.
- Which to pick, Abstraction (ARC): As of Oct 14, 2026, GPT-6 Astra leads Qwen3 235B A22b Instruct 2507 in Abstraction (ARC) at #1 against #82.

As of Oct 14, 2026, GPT-6 Astra is state of the art in Math (since Sep 28, 2026; sources Humanity's Last Exam; methodology v1.0).

As of Oct 14, 2026, Qwen3 235B A22b Instruct 2507 is a former state-of-the-art holder in General LLM (since Sep 18, 2025; sources LMArena Text (style-controlled, overall); methodology v1.0).