Video summary
I Tested Astra 6.2 vs Fable 5.2 (The Astra Hate Is Real)
Main summary
Key takeaways
Summary
The video compares Astra 6.2 with Fable 5.2 across five coding and creative tasks, aiming to test whether criticism of Astra’s quality and “laziness” is justified. The narrator finds that benchmark scores suggest broadly similar capabilities, while Astra is generally much faster and cheaper. In hands-on tests, however, Fable usually produces more complete and polished work, especially when given open-ended prompts.
Benchmarks and efficiency
- Independent results from Artificial Analysis showed close overall performance: Fable scored 64.6 on the intelligence index, compared with Astra’s 62.1. Astra won some tests, including a programming test where it led by 4.7 points; Fable won slightly more tests overall.
- In OpenAI-published figures, Astra led on Agents Last Exam (78.3 vs. 72.8) and was substantially cheaper in that test ($2.89 vs. $60.50). Fable led on Frontier Code 2.0 (49.1 vs. 44.2).
- Across the video’s tasks, Astra consistently used less time, money, and fewer tokens. The narrator’s central question is whether those savings matter if the result needs substantial improvement.
Five practical tests
-
Game design: “Three little pigs defend their homes from wolves”
- Astra produced a basic 2D point-and-click game and stopped after about 28 minutes, using roughly 2.6 million tokens and costing $2.66.
- Fable spent about 1 hour and 53 minutes, used around ten times as many tokens, and cost about ten times as much. It created a more developed 3D-style game with multiple pigs, distinct weapons, upgrades, settings, and substantially more generated images.
- The narrator considered this a clear win for Fable: Astra’s output seemed far short of the requested AAA-style game.
-
Web design: Fictional freelancer accounting app
- Astra’s site was described as generic and visually mediocre.
- Fable’s design included more motion, animations, and small visual details, though some interactive elements did not do much.
- Astra finished faster—about 13 minutes versus 52 minutes—and was cheaper, but the narrator judged Fable’s result much stronger.
-
Interactive 3D turbofan explainer
- Astra built an explorable engine with zoom, component selection, airflow tracking, and explanatory text. The narrator considered it acceptable, but not especially impressive.
- Fable offered more views and controls, including cutaway and X-ray views, disassembly, airflow visualization, and changing engine statistics such as temperature, pressure, and speed.
- Fable was judged the clear winner in depth and presentation. The narrator felt Astra’s result looked underdeveloped given its expected strengths in 3D work.
-
JavaScript-animated explainer video
- Both models generated narrated videos explaining how language models are trained.
- Astra’s was considered solid but comparatively restrained, with weaker voice acting and a more uniform visual style.
- Fable’s felt more energetic, varied, and dynamic, so the narrator preferred it.
- Fable also spent substantially more time and resources: roughly 1 hour and 53 minutes on the initial version, compared with about 17 minutes for Astra, with further time spent adding voice acting.
-
Video editing with Hyperframes
- Each model edited the same raw 30-second video segment. Astra’s version was considered good and comparable to the version used in the narrator’s own video. Fable’s was also strong and slightly preferred, though the difference was subjective.
- Unlike in the earlier tasks, the narrator supplied more style guidance and source material here. Astra’s result improved with that added direction.
- Astra was faster and cheaper—about 48 minutes and $7, compared with Fable’s 1 hour and 34 minutes and $11.50—so the narrator counted this as Astra’s win.
Overall conclusion
Across the five tasks, the narrator’s combined figures were about 2.5 hours and $18.11 for Astra, versus 7.5 hours and $86.63 for Fable. Despite that efficiency advantage, Astra was judged weaker in all but the video-editing task.
The narrator’s main criticism is that Astra tends to do the minimum when given a broad goal and room for creative judgment. It performs better when given clearer instructions, examples, and constraints. Fable, by contrast, was more likely to develop an open-ended idea into a detailed result. The narrator concludes that this is disappointing for a flagship model expected to handle ambiguous, high-level tasks—not just follow tightly specified steps.
The video is a comparative review and practical test, rather than a step-by-step tutorial. It also briefly promotes the narrator’s updated Claude Code Masterclass.
Main speakers and sources
- Main speaker: The video’s narrator, presenting tests and opinions as the reviewer.
- Benchmark sources referenced: Artificial Analysis and OpenAI-published benchmark results.
- Prompt inspiration: Nate Perritt of Anthropic, whose earlier game-design prompt inspired the first test.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.