GPT-6 Astra in Practice: Tests, Use Cases, and Feedback

A poster says GPT-6 Astra has become slower and dumber because of its high price and heavy usage, adds that their own 5.6 Sol is also poor, and suggests switching to the newly released Claude model for now.

The author argues that GPT-6 Astra's official release materials contradict the claim that power is not an AI bottleneck, citing the largest training run in company history, over 100,000 GPUs, and a limited initial rollout due to compute constraints.

Astra scored 88% on Serafim's reasoning-induction benchmark, where Fable 5.1 previously managed only 33%, a result the poster calls near-saturation of the benchmark.

The author says the review shows GPT-6 Astra used for real work rather than game demos, and links to the video on YouTube.

A prompt template asks the model its knowledge cutoff: answers of 2024 suggest GPT-4o or 4.1 mini, December 2025 suggests a downgraded model, and real Astra says it is unsure.

The author shows, in a video, how much of the weekly usage limit was consumed after running GPT-6 Astra at the Medium setting for a week.

On the same physics-animation prompt, Astra was fast and light on quota but looked simple, while Claude looked more realistic but used the full 5-hour window.

Using Onshape and a custom plugin it wrote itself, GPT-6 Astra worked on rover parts, and the results got more impressive the harder the author pushed it.

The author argues that the Grok bot, the Claude update and GPT-6 Astra's recent benchmark results are changing moat dynamics and how much and how tools are used.

A Japanese user had Astra operate Autodesk Fusion through MCP to design a children's lathe toy from concept to structural design, and also generate an explainer video.

The poster asks for real examples of people earning more money than before with GPT-6 Astra, and so far there are only questions, no answers.

The author embedded Claude Code inside Codex so Astra writes a rate limiter and Claude flags issues; Astra scores strongly but still has blind spots.