GPT-5.6 Max beat Fable 5 on agentic workflows.
Not as relevant now :-(: I had an opportunity to deeply test both Fable 5 and GPT-5.6 Max. 5.6 is clearly better than Opus 4.8 at everything (slightly faster, too, though that depends on the load). Vis-a-vie Fable, it is clearly worse on coding, but better on agentic workloads. I had Fable write code, 5.6 run experiments - dreamy…
Positive users appreciate the benchmark details comparing GPT-5.6 performance against Opus 4.8 and Fable 5, while negative users complain that the models remain inaccessible due to release and policy barriers.
No Digg Deeper questions have been answered for this story yet.
@MParakhin Now tell us how in the world you get access to both!?
@veeransg5 No, ultra = workflows on high, the trick is to use max and then say: "Please start workflows with multiple agents..." - agents inherit max effort, makes a big difference.
@AlanRBlair I did. On agentic non-coding (taking actions) I found 5.6 clearly better. On discussing history/math Fable has an edge.
@manabiSRS Ultra is just multiple agents, but each lower reasoning effort - high. You can start them from Max - then they inherit Max, makes a difference.
@Khalin_George Oh yeah, endless critique loop. Fable is very good, so, maybe only 3 iterations. 5.6 makes far fewer bugs than 4.8, but is no Fable - so, 7-8, even 9 iterations "review changes, find bugs - fix - review changes, find bugs - fix, ..."
@SebastianSzturo Testing a preview. Don't have access to either anymore :-(
@MParakhin Tell us more! No one else has tested both and spoken about it!
Any chance you used both for non-coding applications?
@MKuliasov “Take these files with results of previous experiments, parse out x, y, z, analyze, figure out what went wrong, prepare new runs, schedule these machines, be careful with the machine X - it is running this other experiment, keep iterating”
@MParakhin Amazing, thank you. Good to know someone can snip at Fable's heels.
The 2 days I used Fable; it did have a bit of that Big Model Smell - same vibes with 5.6? Or just an improvement in 5.5?
@MParakhin What's the feeling of working with GPT-5.6 comapring to Fable? I hate talking with LLMs outside of just work, but Fable, even in conversations about coding, gave a nice feeling of working with something that is capable, understands nuance, and is even empathetic.
@david_saint_ 4.8 is better, these tests are all well documented: https://toloka.ai/arena
@MParakhin What kind of setup do you have for running experiments? Does it involve extra agents for verification?
@MParakhin Did you try 5.6 Sol Ultra?
@RiskProMax https://toloka.ai/arena
@MParakhin Ultra and the max reasoning mode is the same?
@tonychang430 https://toloka.ai/arena, but no 5.6.
@kgonia7 Fable probably feels a bit smarter than 5.6…
@MParakhin @AlanRBlair So they conclusion as always is to have both and use them at whatever they are better at
@MParakhin fable for code, 5.6 for agentic work, that's the barbell most people haven't figured out yet and now one half of it is banned lol
@SebastianSzturo @MParakhin shopify cto...
