OpenAI May Skip GPT‑5.x and Launch GPT‑6 Within a Month to Counter Anthropic’s Mythos

According to rumors from social‑media leakers Leo and Andrew Curran, OpenAI intends to make GPT‑5.6 the final model of the 5.x series and jump directly to GPT‑6, potentially within a month of the GPT‑5.6 public release.

Timeline :

June 26 – limited preview of GPT‑5.6 begins.

July 9 – GPT‑5.6 becomes publicly available.

Within the next month (by late July) – GPT‑6 may be launched.

Technical route change :

GPT‑5.5 and 5.6 are built on the "Spud" foundation (~4 trillion tokens).

OpenAI originally planned to extend Spud to GPT‑6.

The plan has shifted to a brand‑new, larger pre‑training base for GPT‑6.

Why the shift? The leak attributes the change to Anthropic’s Mythos model. Curran notes that Mythos triggered a chain reaction, including a U.S. government export‑control order that temporarily halted external access to Mythos, prompting OpenAI to respond with a larger model.

Other players’ actions :

xAI : training a 10 trillion‑parameter Grok model; earlier versions (1.5 T) already completed, with 6 T and 10 T versions running in parallel on Colossus 2; Grok 4.5 tested internally at SpaceX and Tesla on June 28.

DeepSeek : plans to release the V4 full version in mid‑July (preview released April 24); will introduce double‑price peak‑hour pricing; rumors claim its capability may match or exceed GLM‑5.2 (released June 13); development of a larger model targeting MiniMax M3 Pro is underway.

MiniMax : M3 Pro has 2.7 trillion parameters; slated for open‑source release in Q3; expected to become China’s largest open‑source model (according to The Information).

Curran also cites internal messages indicating that both OpenAI and Anthropic see rapid capability growth, stating, "nothing above us but air. No ceiling," reflecting strong confidence in continued progress.

From publicly observable trends:

Model release cadence is accelerating (GPT‑5.6 moved from internal testing to public in two weeks).

Mythos has indeed provoked government‑level intervention.

All major labs publicly announce plans for larger‑scale training.

Outlook for the second half of 2026: U.S. labs, spurred by Mythos, are shifting to substantially larger pre‑training regimes, while Chinese labs accelerate open‑source efforts, with GLM‑5.2 and DeepSeek V4 approaching the performance frontier of closed‑source models.

Note: the information originates from unverified social‑media sources and should be treated as rumor.