OpenAI Unveils GPT-6 Astra with Record-Breaking Token Count

OpenAI has released GPT-6 Astra, a significant update to its language model. The new version boasts over 1 million input tokens and five reasoning levels, which can result in higher per-token pricing but lead to efficiency gains and lower total task costs.

Benchmark data on ARC AGI 3 shows that this approach can be effective, reducing overall costs. This is not the only update in the field; Anthropic has released Claude Fable 5.1, which matches leading scores on the Artificial Analysis Intelligence Index v4.3 and eases cybersecurity limits while reducing verbosity.

Other major players like Google, Meta, and Microsoft have also pushed new speech-to-text models that achieve under 4% word error rates. Furthermore, Johns Hopkins and Apple have introduced SelfCompact for rubric-based LLM context pruning.