After Anthropic launched Claude Fable 5 and Mythos 5 on June 9, power users at SemiAnalysis testing the models ran into frequent nonsensical refusals that prompted them to try OpenAI's Codex instead, producing a visible shift in daily token share away from Claude's API offering.
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
Several testers who switched during the trial period reported preferring Codex over the new Claude variants, underscoring how easily usage can migrate when friction stays low.
Anthropic notes the classifiers trigger on average in under 5 percent of sessions and can fall back to prior models, yet the frequency of harmless refusals in real workloads stays unclear beyond the initial SemiAnalysis observations.
Many users prefer OpenAI Codex over Anthropic Opus after model refusals because it executes without second-guessing or paternalism, while some call Codex unreliable or trash.
I will be honest:
1. Since Opus 4.7 I get pushback in unexpected ways from Anthropic models
2. GPT-5.5 doesn't do this
When I know what I want to do, I prefer #2. I don't like a vendor knowing better what I do, how I do it, with zero transparency. I humor Opus but like GPT...
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
Is AI via API calls a good business with so little if any loyalty or stickiness?
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
@VijarKohli Nah the model is materially worse still outside multi-modality
@dylan522p Cannot believe it took you guys so long to realize GPT-5.5 in codex is a better experience than Opus 4.8
@dylan522p No love for Gemini?
@dylan522p Codex is significantly better than CC
👀
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
@dylan522p @zephyr_z9 Wait the split is this bad? Codex has been better for a while
@dylan522p Codex 5.2+ is legit significantly better for real proejcts than Opus. 5.5 is WIP.
I found Fable crushed a project Codex was stuck on for weeks tho, with GPT pro also failing to unstick.
OpenAI seems to be limiting the near unlimited Pro compute finally, too xD
@dylan522p @zephyr_z9 Bro just @ Dario at this point lmfao
@dylan522p we know who is most annoyed
@dylan522p 4.8 is kind of weird. 4.7 is better. 4.8 has been a complainer for me.
@dylan522p @zephyr_z9 • OpenAI deserves credit • Gemini should get recovery therapy • Anthropic needs to stop being misanthropic
@dylan522p This is what happens when refusals break workflows
frustration compounds and people actually try the alternative
@dylan522p Fable is objectively the best model, but GPT-5.5 is my fav as a daily driver. Opus and Sonnet aren't worth the cost. I'm very excited for the OpenAI release that is coming soon.
@dylan522p Anthropic still just dominates. I know gpt 5.5 is good. But my finance agent still uses opus 4.7... Why? I spent most of my time tuned the harness to fit Opus... But for long term, I will have to spend more time to tune GPT harness.
@dylan522p OAI and the Codex team can do the funniest thing next week, I expect they will.
@dylan522p @zephyr_z9 I think I see a one pixel wide blue line at the top
@dylan522p Wait till they discover 5.5 pro extended for analysis work...
After Anthropic launched Claude Fable 5 and Mythos 5 on June 9, power users at SemiAnalysis testing the models ran into frequent nonsensical refusals that prompted them to try OpenAI's Codex instead, producing a visible shift in daily token share away from Claude's API offering.
Usage share of OpenAI grew vs Anthropic yesterday despite Mythos 5 / Fable 5 launch Multiple power users at SemiAnalysis tried Mythos / Fable Got refusals for nonsensical reasons Got pissed off at Anthropic Gave Codex a legitimate try Now they actually prefer it to 4.8 Opus
