Gemini 4 Pro / Flash: benchmarks, hands-on tests and leaks

The author feeds an archive of past tweets to Gemini 4 Argon to clean into atomic JSON, then derives topic-selection and writing skills for batch-reproducing their own style.

A Google AI Pro user finds Gemini 4 Argon's Medium thinking level sufficient for most tasks, with High prone to stalling, and suggests Google add an auto mode that picks thinking intensity.

With Gemini 4 pushing concept generation to single-digit dollars, the author proposes a five-step workflow: spread 40 directions, cull derivatives, pair survivors, push the winner, then lock.

The author argues better AI text comes from using well-tuned models rather than prompting weak ones, citing Claude Opus 5.5, Haiku 5.5 and Gemini 3.8 Flash / 4 Argon.

The author shares a set of instructions that can be added to AGENTS.md to reduce sycophantic responses from Gemini 4 Argon, linking to the original thread.

A roundup of refreshed GitHub jailbreak prompt repositories, including Spiritual-Spell-Red-Teaming with new prompts aimed at Opus 5.5, GPT-6 Sol, Gemini 4 Argon and Grok 4.7.

The author suggests letting Gemini 4 Argon and 3.8 Flash handle big-picture planning, then sending results to bundled Opus/Sonnet 5.5 or external models for review to offset hallucinations and sloppiness.

Citing Anthropic data that manual step review catches only 14% of errors, the post shares a three-step method using Opus 5.5 or Gemini 4 Argon as an independent QA chat.

The author shares a full-codebase security audit prompt, recommends a top reasoning model, and says Gemini Argon 4 is notably sharp at it.

A tip for using Gemini 4 Argon to export an image's colors, typography and composition as reusable JSON style data.

The author explains potato mode as one command that fetches only the skill a task needs, like an encyclopedia index; on stream it fanned out to four helpers and included Gemini 4 Argon.

The author says Gemini 4 Argon wrote the paper's first draft and figures, then rewrote it in his style after being fed his past first-author papers.