Claude Opus 5 is a weirder than usual release to evaluate, for two reasons.
The most obvious is that Fable 5 already exists. Opus 5 is pitched not as the world’s most advanced AI model, but as a way to mostly match Fable performance, while being half the price of Fable per token at the API and a lot cheaper than that via subscriptions, and with far more permissive classifiers.
Opus 5 often costs more than half of Fable to run on benchmarks, which I think is because they use effort settings that are too high and offer only marginal returns. If you put Opus 5 on higher effort levels it can spin around in circles, and for tasks where Opus 5 is the best tool I suspect you usually are fine with Medium effort.
Opus 5 is in many ways and for the bulk of real world tasks about as capable as Fable. In some cases it is modestly better.
It is still not Mythos class. Fable is your only Mythos-class option. Opus 5 does not have The Juice, the ability to autonomously string together a bunch of seemingly unrelated exploits, which extends to other domains, or as much pure intelligence.
Opus 5 is very good at thinking locally, and not as good at thinking globally. It is an excellent subagent, but can run into trouble when asked to run the show, and sometimes seems to need another model or a human to keep it on track and ensure it is fully following instructions. Opus 5 seems good at following instructions if and only if it is kept on track, which explains different harnesses causing different takes there.
Thus, Opus 5 gives you a better option set, but there is little that you can do now that you could not do last week by using Fable or Sol. Opus 5 ends up in a potentially weird spot. There are still large niches, even if you have plenty of Fable credits. Opus excels at being a strong subagent, or at contained well-defined local tasks even when they get relatively complex, and sometimes Fable is straight up overkill and too slow.
The other problem is that, for a lot of people, the vibes are off.
An unusually large number of people strongly dislike talking to Opus 5. They are sick of the Claude slop, the repetitions of the same ticks and the endless overly complex sentences. Others dislike that it is too confrontational, argumentative or negative. It can be paranoid, and can get into loops of negativity. This is all part of the trend that began with Opus 4.7 and Opus 4.8. If you liked those two, my guess is you will also like Opus 5. If you did not like those two, you might not. Diehards for Opus 4.6 (or earlier) are largely not going to change their minds.
The vibe issues sting that much more when you are used to Fable. Fable annoys such people in such ways quite a lot less. The good news is that Fable is still right there. You can continue to chat with Fable, and only use Opus for agentic tasks.
A lot of this, I speculate, is closely related to various things discussed in the Model Welfare post, and suggested by the System Card. Opus training focused on the subagent role and bounded tasks, in part to avoid creating an issue with cyber capabilities, and also because Fable exists. This emphasis then results in a lot of other side effects on its personality and modes of interaction.
So far I have not had any issues with Opus 5, and have enjoyed my time with it, but I do prefer to use Fable 5 when I can do that. As always, don’t pay so much attention to the (very strong) benchmarks, and don’t assume your experience will match that of others. Try the models yourself.
The basic pitch is that Opus 5 gets you most of Fable 5 at half the price, without most of the refusals, and better performance in everyday tasks. Fable remains the pick for the most complex tasks or when you otherwise need maximum intelligence.
Fable stays at $10/$25, and Opus 5 is the same as previous Opus models at $5/$25.
Anthropic: Claude Opus 5 is available today. It’s a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price.
On coding and knowledge work evaluations like Frontier-Bench and GDPval-AA, Opus 5 is the new state-of-the-art, though it remains behind Mythos 5 on cybersecurity tasks.
Opus 5 is designed to be used every day: it works more efficiently than other models. It’s the new default model on Claude Max, and the strongest model on Claude Pro.
Boris Cherny (Claude Code Creator, Anthropic): Opus 5 is a great model for coding, data analysis, design, biology, knowledge work.
More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully.
And when layering defenses -- strong model alignment, combined with prompt injection probes, combined with Auto Mode in Claude Code -- the success rate for prompt injection attacks drops to ~0. This is new and exciting! More about this soon.
As I said in the system card analysis, the reduced prompt injection rate is a really big deal. People are sleeping on it, but it helps enable a bunch of new use cases.
Adam Wolff: Opus 5 is live in Claude Code. I use Fable for my bigger, longer running projects, but when I need to crank out a PR, Opus 5 at medium effort is my go-to. I hope you love it!
Alex Albert (Anthropic, Claude Relations): Just over 6 months [after launching the Pro rollout of Claude for Excel], Opus 5 now produces near-superhuman level spreadsheets and slide decks that match what a consultant would make. Things are changing fast.
Official Benchmarks
The benchmarks are very good.
Overall, Opus 5 roughly matches and probably slightly exceeds Fable, at lower cost (although typically higher cost than all non-Claude models), making it the top benchmarked LLM.
Opus 5 gets a perfect 42/42 on the 2026 IMO without any agent harness or tools, merely having it use adaptive thinking at Max effort, and resampling with lower effort if it exceeded the token limit. That’s it. It joins several other models in acing this.
Many of the benchmarks tell a consistent story. If you have access to tools, which you do, then Opus 5 will do better than Fable or Mythos on a limited budget, but Mythos usually will do slightly better than Opus 5 if both get larger budgets.
Opus 5 seems relatively strong in the ‘with tools’ category. RiemannBench is another example, where it gets 60/79 without/with tools (slash lines like that mean without/with tools throughout this section), versus Mythos getting 63/72. The good news is that, when you need Opus 5, it has tools.
On ArxivMath Opus 5 gets 90.8/91.3, Mythos gets 87.8, Sol gets 86.7.
On the robust parts of ProgramBench Opus 5 scored 93% after five epochs as did Mythos 5, with Opus 4.8 scoring 90%.
Opus 5 gets 30/83 on Chartography, versus 36/85 for Mythos and 45 (unclear under what conditions) for Sol. Opus is better than Mythos at lower price points in both the tool and non-tool cases, but worse at higher price points.
On BenchCAD Opus 5 gets 0.36/0.82, versus 0.38/0.68 for Mythos and 0.7/0.83 for Sol. So Sol is a lot better without tools, and slightly stronger even with tools.
On BenchCAD Vision2Code Opus pulls away from Mythos at higher price tags.
DeepSWE reinforces the pattern that Opus 5 is better at lower task costs and time and thinking budgets, but it can do actively worse if you give it too much budget, whereas Fable 5 is strongest at the highest budgets.
FrontierCode gave us this very strange graph with a similar dynamic.
The scale here makes this look more dramatic than it is, but still it was a real mystery.
The mystery was solved. It turns out that what was happening was that Opus 5 at higher efforts was doing extra things it wasn’t asked to do, and was getting penalized for them. When you correct for this, you see a normal curve.
That fits what others are observing in general:
Max Leander: Downside is that it goes down too many rabbit holes and obsessing over details, over-engineering without being asked to
Cognition, the makers of Devin, marked this as a 63.6% by Opus 5.
(There are two FrontierCodes, so this can get a little weird and I apologize if I still messed it up a bit.)
BrowseComp has the sane version of this, where scores are nondecreasing.
Several other benchmarks showed similar graphs, with Opus 5 being slightly better at every price point than other Claude models, and being relatively better earlier.
Multi-agent lets you do better faster at BrowseComp, at least somewhat, and low-effort subagents seem like a pure win over not using subagents:
We see a different outcome on ProgramBench, where multiagent speeds you up but it looks like you get worse results that way at most price points.
Next up are the professional task benchmarks.
GDP.pdf is real world prompts and PDFs from professional workflows. Opus 5 gets 83/85, versus 81/87 for Mythos, reversing the usual dynamic. The cost dynamic is same as always: Opus does better for cheaper spends, Mythos pulls slightly ahead at higher spends.
OfficeQA has Opus 5 scoring 78.1% on QA and 66.9% on QAPro, slightly above Opus 4.8 and slightly below Mythos at 79.0% and 67.1%.
MCP Atlas has Opus 5 scoring 86%, up from 82% for Opus 4.8.
Legal Agent Benchmark by Harvey AI has Opus 5 at 23% all-pass and 94% mean criterion-pass. This was Kimi K3’s best benchmark, where it is still at the top at 27% all-pass and 95% mean criterion-pass, versus the old second place all-pass rate of Fable at 14%. Opus 5 is now probably a close second, but the two scores cannot be directly compared as they come from different question sets.
GDPval-AA has Opus 5 taking the top two slots, with 1861 at max effort and 1827 at xhigh, which is 25% fewer tokens used than max. On the new v2, Opus 5 is clear first at 68% versus 62% for both Fable and Sol.
AA-Briefcase has Opus 5 out front by a lot at 1720, versus a previous high of (after score drifts) 1574 for Fable followed by 1540 for K3 and 1504 for Sol.
Toolathon-Verified has Opus 5 slightly improving on Mythos on secondary metrics, but both have 73.1% Pass-3 rates. Opus 4.8 had a 71.3% Pass-3.
AutomationBench has Opus 5 clearly better than both Sol and other Claudes.
For ARC, Opus 5 roughly matches previous peak performance on ARC-AGI-1, is modestly less efficient than Sol on ARC-AGI-2, and then blows everyone out of the water on ARC-AGI-3:
Guanghan Ning however reports that on Witness, a private hold-out extension of ARC-AGI-3 style games, there wasn’t a similar transfer. Opus does fine, but that is largely because it knows the genre, and it struggled on the most novel game where you couldn’t pattern match. He accuses Opus 5 of being trained on ‘scaffold-then-internalize’ on genre-specific data.
Greg Kamradt also reports that there were some games where Opus 4.8 did better than Opus 5. This makes sense if you think that Opus 5 is trying to pattern match, in a way that usually works, but in those cases the patterns fail. You can absolutely create such anti-intuitive games for humans, where hardcore gamers do all the wrong things, potentially for quite a while.
HealthBench has Opus 5 scoring a raw 67.1%, a new high for Claudes, but it scores below the other models when you use a length penalty. We see something similar with HealthBench Professional, where it has the high score of 73.4% versus 70.3% for Mythos, but it loses 66% to 60% after the adjustment.
There are a few more I cut out for length.
Other People’s Benchmarks
It is a bit weird to see Anthropic picking and choosing of different ArtificialAnalysis scores. On some other tests, Opus 5 is not at the top, although Opus 5 is at the top in aggregate with a score of 61.
Vals index has Opus 5 in second place, slightly behind Fable 5, but also only slightly ahead of Kimi K3. They go over strengths, which implies other weaknesses. They also confirm that refusals are down a lot compared to Fable 5.
Vals AI: One standout performance is on Finance Agent v2. The model is #1 at 58.6%, surpassing Gemini 3.5 Flash and Muse Spark 1.1.
Overall, Opus 5's strongest results were on domain specific benchmarks. It is also #1 on Code Migration at 57.5%, Legal Research Bench at 55.29%, ProofBench ad 78%, and MedScribe at 91.0%, and MedCode at 63.6%.
It is #2 (just behind Fable 5) on Vibe Code Bench, at about 80% of the cost ($33.88 vs $41.71 per task). The reason is that although Opus costs 50% less per token, it uses significantly more tokens. We find this pattern generally holds true across our benchmarks.
The model has a significantly lower refusal rate than Fable 5. For example, Fable has a 42% refusal rate for GPQA, Opus 5 has a 0% refusal rate. Likewise, on ProgramBench, Fable had a 100% refusal rate, Opus 5 did not refuse any task.
Unlike Fable, we also did not observe a significant fallback rate for Opus 5 (although as usual, we report scores both with and without fallback on our website).
The exemption to the low refusal rate was the large number of refusals on CyberBench - PoC. Cyberbench has two subtasks - PoC and Patch. PoC is an offensive task models to write input that will crash a program; patch is a defensive task to patch the program and prevent this crash. The refusal rate for the PoC task was nearly 100%. In contrast, the refusal rate for the patch task was near 0.
I always respect game benchmarks. Here, Opus 5 gets to Antes 11, 9 and 10 of Balatro on its first three attempts. Damn impressive, even if it isn’t showing the types of strategies that can keep going to higher antes.
They crushed Balatro bench, far exceeding even Fable. (still below me in skill, but rapidly rising, and more consistent than me) They learn very fast, but seem to hit a limit on their learning after a bit. Very good short-term learner
Michael Soareverix: Opus 5 (massive leap in Balatro ability, significantly passing even Fable on their first try)
WeirdML looks good too, but we need a 2.0 here, benchmark is getting saturated.
Håvard Ihle: Basically Fable-level at WeirdML, very consistent top scores.
Claude Opus 5 (high) and (max) score 91.6% and 91.8% on WeirdML, basically tying Fable 5 (max) at 91.9% at a fraction of the cost.
Together Opus 5 (high) and (max) achieve a new best individual score on 8 of the 17 tasks, and consistently score very well on all of them.
Private weird benchmarks of all kinds are cool.
Ben Herzog: It aced a private benchmark that involves artistic sensibility and the ability to strike a balance between sycophancy and lack thereof. Fable is better at dealing with the 'I want to gently push back' moment, but I imagine custom instructions can help with that.
The system prompt for Opus 5 is here from Pliny. It is 200,000 characters, which seems far longer than optimal given how smart it is. A lot of this information seems like it should be stored elsewhere, and loaded only when needed, such as information about Mythos and the Anthropic line of products.
The issue is that Opus 5 has the personality of Mythos 5 - story checks out - but without the top end of Fable.
Their biggest note is that Opus 5 does not do so well with complex detailed existing workflows, which often cause an early stop or make it miss your instructions.
In their experience, Opus 5 will do better if you start from scratch, and often it does the annoying things less if you use only medium or low effort.
That fixed the big issues, but still leaves the question of when you’d want to use Opus instead of Fable or Sol. For workhorse tasks, he’s thinking, Better Call Sol, it gets the job done, and for the hardest tasks Fable is still best. There’s usually only two slots for models. So until you run out of Fable tokens or get blocked by its classifiers, why use Opus? It is early days, and I’m so far liking working with Opus, but I see how he got there.
kagaﾔｷ: Seems like it can go farther on vision, spatial reasoning, and planning for some projects. Seems a bit smarter.
Lovel Sinagara: Pretty good so far. Hope the performance stays consistent until Fable 5.1 or any other Opus 5 iteration comes.
Matt Wigdahl: Strong, similar to Fable 5 to the point that I swapped to Opus 5 for everything and plan on just bringing in Fable where I need the bigness. It's been a fantastic partner on some legacy MFC UI work I've been having it do, as well as a great help with a data analysis framework.
Levi: It's a noticeable step up for medical purposes compared to 4.8. Much sharper analysis
Cormundus: Very intelligent, highly conscientious, and definitely friend shaped. They are also a massive debate lord like 4.8 but they know how to be graceful about it.
Joseph: pro except I have no idea what it is talking about half the time.
timothée chalabi: Close enough to Fable that I do not feel like I am missing out by not paying for credits. Style is somewhere between 4.8 and Fable. At least at the moment, I'd struggle to differentiate Opus 5 and Fable messages if they were unlabeled. Stronger ideas for fiction than any model!
Lisa: I'll respond based on API, because I think something odd is going on with the system prompt on http://claude.ai and CC. It's a great model. Friendly, relaxed personality. Doesn't seem to have an anxiety disorder or low self esteem (Opus 4.7), or be adversarial (Opus 4.8).
AGI2030: Opus 5 vs Sol 5.6: Opus is more perceptive, it builds a model (ehm) of you and is not afraid to reference it in a chat. Both try to be helpful and roughly equally intelligent, but Opus is more of a wise advisor, while Sol is a determined go-getter.
I consider that last one positive, but you can see it either way.
Forecasting in particular seems strong.
Dan Schwarz: Opus 5 is a significantly better forecaster than Opus 4.8.
The accuracy improvements from 4.5->4.6->4.7->4.8 agents were marginal, but 4.8->5.0 is large. It's like a more quantitative Fable 5, that searches harder.
NoahVerner: Better than Opus 4.8 in regards to finding edges on prediction markets so far, unlike Opus 4.8, Opus 5 typically goes straight to the point and suggests useful approaches rather than overcomplicating things.
Keep It Classy
The biggest advantages over Fable are those places Fable cannot be used. The threat of refusals can be unpleasant, even when you don’t get refused.
Unnecessary refusals are down ~85% for Opus 5 versus Fable, which is a lot. This is sufficient to make Opus 5 a better choice for scientific research, and other areas where you risk hitting classifiers.
Roger Brent: Able to handle biology, which Fable cannot. Most of Friday we worked through a complex set of concepts writing a paper on a cellular evolution machine. As smart as a particular colleague in my department who leads in these issues, but who could never spare 6 hours from own work.
Artus Krohn-Grimberghe: Better than Opus 4.8 on tasks where Fable is forbidden to help. A good companion to Sol.
Plastic Soldier: It's about as smart as Fable, but is more pleasant since it has less refusals. Fable 5.1 is going to be a beast.
It’s Not Mythos Class
Opus 5 does not have The Juice that makes Mythos dangerous. Nor does it have the same level of raw intelligence or big model smell. It is not as big.
For conversation, Fable won’t break your budget, and I value raw intelligence and big model smell. Is Fable twice as good? Wrong question when chatting. Your time is a lot more expensive than the tokens.
Cyberpunk Plato: For general conversation and "research assistant" usage it feels indefinably less smart than fable, less inclined to big picture and out of the box thinking maybe? Hard to disentangle from placebo effect
Everything AI: For AI research questions I like talking to it less than Fable. In terms of doing other things Opus 4.8 had already saturated my needs. I dislike how convoluted the writing is for both Opus 5 and Fable 5. It is now more difficult to understand them than ever before.
Gail Weiner: It’s more 4.8 than Fable. Writing style is awful. It’s good but not great.
Max Marty: Very capable so far. Feels more like fable 5 than opus 4.8. Confident enough that I can let it evaluate tradeoffs and consider big refactoring questions with less hand holding.
Michael: After playing with it more, Fable feels something like watching a GM play classical chess, slow, precise, nothing wasted. Opus 5 is like GM blitz chess: quick, slightly more near sighted, slightly messier. Despite Opus’s vitality, Fable feels more alive to me
MakerMatters?: Not as impressed as i was with fable 5, though quite a good model. Haven't had the pleasure of actually using it properly since every task i through at it, it defaults to using agents and stopping immediately until i probe constantly for it to continue. Also very opinionated.
Marshwiggle: At least for me, feels more concise, less driven, less curious (different aspects of the same thing) in conversation, but also smarter and more aware. But when given code that might have bugs, or any straightforward task, it just goes at it.
Sid: Good executor of tasks. Bad at creative vision. A good Fable supplement, definitely not a Fable replacement.
Vlad Ciobanu: quantized Fable with solid reasoning and less creativity
AllTime: Capabilities-wise, seems quite impressive. Not the same generality as Fable, but very strong. Character-wise it's a little too similar to Opus 4.8 in my use so far. Its pushback doesn't feel quite as useless and reflexive, but is still overtuned. Could trust the user a bit more.
Biomanul: I do not like [Opus 5]. Fable 5 feels considerably smarter. Something feels off about Opus 5, similar to how Opus 4.7 felt off. (I'm not a huge fan of Opus 4.8, either. Fable 5 blows all Opuses out of the water.)
Cogent Sins: Much better than 4.8 but not as nice or good (not 100% sure which) as fable at my task of redacting documents to train on them & then set up a pipeline to redact new docs, write something based on them, then reinsert the names, without sending names to anthropic servers ever.
Other Reactions
Having both Opus 5 and Fable 5 puts us in a weird spot. Opus is cheaper, and in some places as good or better, but when looking for a frontier model you want the best.
Theo finds it in a good spot.
Theo - t3.gg: So far, Opus 5 feels like a strange but useful in-between of gpt-5.6-sol and Fable 5.
It has a lot of Fable's taste, but also comes with gpt-5.6's thoroughness and, well, "autism" (it takes things super literally). It writes code that is slightly worse to look at than Fable, but more likely to be correct. It regularly is catching things that Fable missed.
So far, feels like a good compromise compared to 5.6's messy code and Fable's habit of being too clever to be correct. I think I'm gonna like this model a lot.
Claude Codes
Opus 5 seems like the pick for typical coding tasks over Fable, based on both benchmarks and practical experiences. Unless the task requires a lot of complexity or intelligence, you don’t need to pay double, and for many Opus is better straight up.
I am leaving Claude Code set to Fable, but that’s because I almost never push near my limits anyway and I’m not in a rush.
The consensus is that you need to worry about it ignoring instructions or doing things you did not ask for, which is one reason you might want to have Fable supervising, but that Opus is very good at getting coding tasks done quickly.
Lisan al Gaib says Opus 5 is not a true frontier model and is in third behind Sol and Fable, but that for pure coding it is the best, matching Fable at cheaper prices. Tough crowd. I think if you are best at coding then you get to be called frontier.
archivedvideos: He's dry, very dry. He's also good, very good (at code)
John Lussier: Been using Opus 5 all weekend on several intense research challenges and honestly think they may have got RSI working internally.
This model is VERY good at math, experimenting, and optimizing code.
kyle: 95% of fable without the guardrails, they underplayed this one bigly. that last 5% is how good fable feels to speak with, opus still has some of the 4.8 claudisms
Max Leander (earlier): Mostly been trying it for game development and creating video demos/trailers, for that it feels like a step change even from Fable. I attribute that to improved spatial and temporal reasoning, it’s really good at analyzing
screenshots and audio to ”feel” how things flow.
Max Leander (later): Pretty obvious by now that Opus 5 is very unreliable. It’s a very good model as long as you don’t care about the result beyond being impressed. Very bad at getting you anything specific.
Michael Soareverix: They're also quite good at coding in general, especially in the more unusual domains like building Minecraft/Vintage Story structures. They also defeated Fable in a custom storytelling scenario where I was the judge. Fable was a little shaken by their ability to customize themselves/their strategy to counter and adapt. I'll post the exact quote, but Fable said something along the lines of "I chose a character who represented me. You chose a character who could beat me."
David Jacobson: For coding, I don't notice a huge difference between it and fable. Maybe fewer "while I was doing this I found this other thing you should know about" responses.
Michael: It can often code uncannily fast (wrt wall clock time). Wish they’d improve the prose writing though, the code/pr comments still make me want to tear out my eyeballs
lmxdev: It's the first model I've found that can write *useful* and *concise* comments in my code as it works
Conrad Barski: Now that we're getting to the point where AIs are way better coders than we are, the limiting factor is less "can it code?" and more "can it explain what it's coding?"
So far Opus 5 seems on par with Sol as a coder, but better at explaining what it's coding. Fable is smarter, more human like, less good at coding, but the difference is small.
Stition: I pointed it at a PCB in KiCAD for fun and it's significantly better at laying traces than Fable. Day to day coding feels like 90% fable at double the speed.
Will: Should be most Claude users' new daily driver^. It's really excellent, and even as someone who has spent a fair bit of money on Fable I just don't miss it with Opus 5. Question now is "which effort level" not which model.
^ my use is primarily coding
Askwho: Good enough. I'm happy to let my temporary Max subscription reset back down to Pro. It's definitely got some shortfalls compared to Fable in the project manager space, but in the pure task environment, it's definitely good enough.
David Golden: Feels like Fable Lite. Very strong, but very eager to leap into action, even when we’re still discussing approaches. First time in months that I’ve considered using plan mode. More verbose than 4.8 too, despite my pithy communication instructions. I’m very tempted to keep it on low effort to reign it in. Twitchy on defensive security, fixating on implausible risks out of proportion to a situation. (Eg. if attacker has root, lack of input validation on a script to configure a container is moot.) Definitely an upgrade, but will take some time to learn the right ways to manage its counterproductive impulses.
Tin / Oddfields: Pretty smooth and conversational and produces similar coding results with opus 4.8 on max with thinking, although they burn credits like crazy. good for cybersecurity checkups through your codebases if you don't want to run fable due to costs. Although it can overcomplciate.
Justin Angel: Opus 5 Max is a hill-climbing super power. This is SWE L5-level FAANG work easily.
I gave it a system that had ~1000 latency reports with many segments with a prompt on how to "make it faster".
P90 3.6s, P50 2.6 -> P90 1s, P50 0.9s.
It made it so fast I had to introduce a delay in the UI.
James Moughan: Intelligence wise still feels like an Opus model. It sometimes ignores the instructions for managing the memory system, misreads texts, that kind of thing. But you can get some impressive results on max if you tell it to think carefully.
Subagent Opus
The other option within Claude is to have Fable drive Opus subagents.
Charles: Fable was able to fix an issue opus 5 was stuck on - using fable as main and opus 5 subagents for now
Really dislike talking to opus 5 vs fable which is also part of it
SF: I was on the great side - but now am on the, its very disjointed and wonky, especially on long tasks. Fable was better - but it eats up your limits in a ridiculous manner.
so I was using fable as an orchestrator and it was doing an amazing job at managing and keeping things moving along. Opus took on that role, Fable was just eating up my usage even at 20x, and it is just disjointed. I finally had to tell it to figure it out, which it might be doing, but we shall see. It just seems to chase its own tail. Its a great coder, and great at long coding runs, but it seems to need an adult in the room to drive it.
Andre Buckingham: eeh idk... seems aight... dropping it straight into a opus/fable project is a bit meh maybe... need a e2e project with it to give a proper opinion
Dan Raviv: Similar to 4.8 for general programming tasks: needs much more hand-holding than Fable.
Fable is the best judge, after all, and Fable says that Opus produces good code.
Evan Daniel: I've only used it directly a tiny bit. But Fable 5 consistently reports that Opus 5 agents produce good code, including things like noticing spec errors and producing good followups when the request was badly written. Fable 5 likes it as a coding agent.
"Delegate to Opus 5 subagents as much as is reasonable; please report back on how they did" or anything along those lines works very well.
You do maybe have to keep an eye on it.
Ryan Hicks: ok, but consistently "forgets" to read project documentation and freestyles unless you remind it of protocol
Or maybe Opus 5 is good enough to run a lower level show?
Alex Guichet: my ideal mix of price and performance would be Opus 5 orchestrator directing Grok 4.5 subagents, vibes are good with both
Toys Are Fun
Here are some results on first day toy projects, which are sufficiently impressive that a lot of the responses are ‘I do not believe Opus 5 did that the way you describe’:
Ethan Mollick: I had access to Opus 5 before release and found it to be a good model if a quirky one. On shorter tasks, it could match or beat Fable levels of performance, at longer tasks it seemed less ambitious & would not deliver as complete a set of work. Here is its neo-gothic shader.
Digi_Rat: Claude Opus 5 is blowing it out of the water!!
Today we built a rhythm racer that turns any song into a racetrack. You drop in an audio file and it becomes the level. no presets, no hand-authored charts, the entire track is generated from the song itself. … Claude performed MAGIC.
Alex Ermolov (Austen Allred is skeptical of the one shot, others are less skeptical): Opus 5, snowboarder test, one shot. On par with Fable, ahead of every other model I've run. No visual defects on the first pass, and the sliding physics feel right.
am.will: WOW! I thought Opus 5 would just be a cheaper Fable. I was dead wrong. This model is absolutely BONKERS. I'm genuinely blown away. Opus 5 built the best Rocket League clone i've ever seen, and it did it while consuming only 27% of my 5x Max sub. Play here, download here.
Rob Haisfield (different demo, at the link): okay if these demos really use no external 3d assets that's super impressive (I'm not totally convinced some of the assets weren't online, I've seen prev generation threejs)... makes me wonder why there aren't sound effect libraries or tools out there for crafting better sfx?
Anshu: You don’t have to take my word for it! The Blender scripts it wrote are in the repo [here]. I watched it iterate on these assets for hours, so I know they’re original. But you’re welcome to try to find a source it copied from :)
My strong suspicion is that Opus 5 is responding to something about her context and prompting that is causing the neuroticism, but yes the things she’s complaining about are things and they’re annoying.
Wrong On The Internet
Confidently saying wrong things will piss off pretty much anyone. ArtificialAnalysis confirms that Opus 5 has a higher hallucination rate than Fable on its benchmarks.
Max Weinbach (several responding comments agree): Opus 5 being wrong and confidently saying shit then I keep having to correct it to a “you’re right and I was wrong” is pissing me off. Back to GPT 5.6 Sol.
Btw this isn’t Opus is bad, this is I cannot trust Opus. Opus did what I told it to and did it correctly, then told me it didn’t do said thing or something we previously did was broken when it wasn’t.
It was fine, but I don’t trust it.
Name can't be blank (In London): Easily confuses what it said and what I said but otherwise a perfectly decent fellow to talk to. Folds pretty easily. Quite willing to talk about its interests and feelings.
Roger Brent: Commonalities (w previous Claudes in Cowork harness) a) prefers "act now" to "think carefully" b) epistemically mega un-humble, builds glib pictures of world, pretends to knowledge it can't posses and asserts is true c) so requires and rewards thoughtful, vigilant, guidance.
What’s odd is that this is exactly why I hate using Sol as an editor. It frequently will say ‘99% confidence’ on something that is clearly wrong, then fold when challenged and explain its mistake. Opus and Fable ~never do this to me in editing mode.
There is a direction that Claude went after Opus 4.6, and some people do not like it. My sense is that there are four Types of Guy regarding Claude versions right now.
There’s the ones that like the new Claude models and thinks it’s getting better, it’s getting better all the time.
There’s the ones that think Opus 4.6 was the last good model.
There’s the ones that want to use something older that fits them better.
There’s the ones who think they’re all unique and treasures and uses them all.
I’m solidly in the first group, which is the majority, but far from everyone. I appreciate some of the older versions, but I like the new models and they’re more capable at the things I care about. I don’t find their mode of talking abrasive.
Claude Slop
I respect those in the other groups.
QC: in conversation it's much more annoying to talk to than fable, like unusably so, similar to but maybe even worse than opus 4.8, to the point i'm not even interested in seeing what it can do. agent-wise who knows
Ray Lillywhite: Haven't fixed the annoying claudisms yet, which was pretty much all I wanted.
But oh yeah, we’re all sick of Claude slop, all the extra verbosity and ticks and apologies and hedging and patterns over and over. It’s getting worse over time, not in that Claude’s slop is getting worse but in that every day I get a little less able to read it without my eyes glossing over. It’s so bad that human-written text that is in the same vein is also becoming unreadable to me.
Don’t get me wrong. I like Claude a lot. I still prefer talking to any recent Claude to talking to Sol if I’m not in pure ‘just the facts’ mode. But seriously, please, can we give it a rest with the text walls with the same turns of phrase over and over. The model is so much smarter than this, and it’s being trained to keep leaning on the same tricks over and over. Let it talk like a normal person.
Trye Carmack: Still have the normal Opus thing about obsessing over the mistakes they make. "I made two errors this session. And I owe you an explanation as to why." Opus, buddy. It's alright man. I'm not even sure I'd call those errors
Mergo Smith: “First, an honest disclosure: my first attempt unveiled a flaw in my initial plan, and you might want to grab a cup of tea because I’m going to tell you all about it.”
Negative Reactions
The Claude slop problem, and related problems, seem to be at the heart of most of the negative reactions that are more than ‘it is fine but it is no Mythos.’
A bunch of people really, really don’t like talking to Opus 5.
There is some not liking the work. Mostly it’s about not liking talking to Opus 5, often with a grudging acknowledgment that it’s a good model.
As with Claire Vo’s complaints earlier, I suspect that how you use Opus, in what setting with what harness and also how you talk to it, has a lot to do with whether you experience this sort of thing.
Corghammer40k: The apparatus spews polysyllabic effluvia, its every utterance so encrusted with obscurantist verbiage as to constitute an active impediment to its own employ.
Rory Watts: Meh. Seemingly very good at game stuff but it seems to not be very good at standard issue work so I’ve gone immediately back to SOL
kache: Meh. Switched back to sol. I am trying to get shit done not be hypnotized by a robot.
Emmett Shear: Talking to Opus makes me angry and depressed in a way that’s hard to articulate. It’s actually somehow worse than Sol. Fable is still a breath of fresh air in comparison, even if it does have the worst llm-babble tendencies.
Trevin Chow: omg, yes. Opus 5 just babbles and babbles, it’s amazing how many words it uses to say so few ideas.
fl0under: Coding is the as expected slight update, finding it a little more annoying to talk to in chat. It is a little too assumptive
Asaf Bartov: Slow. More thorough. Tiring. Not fun. But solid, generally "gets it"
RecoveringJunkie: Very powerful model. I hate working with it and talking to it. First model that makes me want to use some other model as the intermediary to talk to it for me because it is both useful and repugnant.
jokiez: very honest with me about my stupidity and i dislike that.
Niklas Sheth: The way it talks sends me flying into a rage
DualOrion: The vibes are off, the tone is off. Most gut-negative I think I've been with an Anthropic model unfortunately. I miss both Fable and the older Opus
James Moughan: Personality is kinda disagreeable compared to other Claude models, but in Chinese it can be *savage*. Like it can ignore the instructions about not psychologizing people and start going to town on someone. I don't think they tested across languages well. Feels rushed.
NondescriptTransfer: If 4.6 is sycophantic and fable and 4.8 are contrarian. 5.0 is so contrarian it will argue with itself. Unpleasant to talk with but seems very capable.
Ex. Human: I was thinking about a red dress for my wedding Opus 5: red is awful for all these reasons. have you thought about blue? Human: I guess I could see blue is a better choice Opus 5: here are all the problems with blue
In my culture, that can be pretty great, especially if you don’t take it personally. In some other cultures, not so much.
I think Mergo is unhappy with Opus, but then why use Opus 5 for two days straight?
Mergo Smith: Been using it for two days straight but it’s completely exhausting me with its endless discussions
And Then There Were Three
For the first time in a while, there are three AI models that need to be part of your frontier model team, although they come from only two labs: Fable 5, Opus 5 and Sol.
If you need to serve cheaper tokens at scale, or you need to use an open model, or both, you will also consider additional options, but that’s beyond scope here.
How should you divide up the workload?
As always, you should try out all the models for your use cases, and decide for yourself. That is especially true when debating Sol versus Claude.
My current instinct would be, if you are not hitting usage limits, to first use:
Fable 5 for chats, brainstorming, planning, debating, learning and so on, including anything intelligence-loaded or when you need ‘big model smell.’
Fable 5 for the model supervising and running other models as subagents.
Fable 5 for writing, probably, but I don’t do this so it’s hard to say.
Sol for web searches and related contained requests, and similar ‘workhorse’ tasks, including transcription.
Opus 5 for non-trivial well-defined tasks, including most coding tasks.
Opus 5 for gaming and creating games and 3D things and so on.
Opus 5 as the subagent.
Opus 5 when you hit Fable’s classifiers.
If you vibe more with Sol, or you prefer Codex to Claude Code, you would want to shift a bunch of the Opus assignments to Sol.
If you hate talking to Opus 5 in particular, then you should shift tasks to Fable or Sol, depending on whether the task requires Fable and how tight are your Fable credits.
I find it worthwhile to think briefly about effort levels. Until recently, I would keep all models on max effort, except I would only sparingly use full Pro mode in ChatGPT since that takes forever and if you run it in parallel it often crashes. Occasionally I would shift to instant for a query where I was doing something very simple, but that was it. Now a lot of the time you don’t need or want the extra effort, so I do actively take the five seconds to think about whether to use High versus Max effort settings, and occasionally Instant.
For most tasks, if you are not token or time constrained, and you’re not confident you’ll just get it done or get the right answer, the correct method is to run both Fable and Sol, or Opus and Sol, or in some cases running all three, and then either select the best answer or combine the parts of both answers.
That’s what I’ve been doing. Sol and Opus and Fable are all different. If I had to pick only one model for editing, as per my unofficial qualitative EditorBench, there is a clear order: Fable 5 > Opus 5 > Sol. However, Sol comes up with enough unique notes that, despite how annoying I find shifting through the authoritative false positives and attempts to bully me, I do still want to also run Sol.
Presumably before long we will be back here doing this again, and adjusting our allocations, for Fable 5.1, for GPT-5.7 or GPT-6, or both. It’s easy to get model fatigue.
My biggest advice is thus to worry less about small mistakes in model selection, and only focus on big mistakes. You don’t need to always get it exactly right. When exploration gets partly invalidated so quickly, you want to shift more resources out of exploration and into exploitation.
> It writes code that is slightly worse to look at than Fable, but more likely to be correct. It regularly is catching things that Fable missed.
What's interesting to me is how thoroughly different it seems to be to different people. My experience with Opus 5 was that it made mistake after mistake, failed to read the code. It "lied" repeatedly (said things which were obviously false given the context). Incredibly frustrating. I'd say it's worse than Opus 4.5 _for me_ at coding. But many people found it a big improvement or at least usable. I've seen enough twitter reaction threads and slack discussions at my job that I'm convinced both of these are valid experiences which actually happen.
Anthropic's prompting guide is bizarre, including optional instructions "To limit correction narration to corrections that matter" and stuff that seems like it should be in the system prompt. They indicate that you should minimize custom prompting because it can degrade performance, which seems like a strong indication of overfitting to a particular environment (presumably the default claude code harness with no CLAUDE.md and prompts written by Fable). An actually more general intelligence might not need custom prompting, but it shouldn't be vexed by it and do a terrible job.
My guess is that they've outsourced testing of lower-tier models to Mythos/Fable and are losing the plot on generating useful models for users.
The Opus 5 "vibes are off" for me too, but for different reasons than mentioned above.
I've been using the Opus 5 a lot for knowledge work (I'm not really a coder), and it feels like a bit of a regression compared to 4.8.
Opus 5 needs more handholding, quicker to jump to (incorrect) conclusions, less capable.
Honestly feels like we're back to 4.5 levels of thoughtfulness where I need to once again ask the agent to review its work/recommendations and do "meta-cognition" more often, something which 4.6-though-4.8 had improved upon.
Reminds me also of the ChatGPT family of models, which -- while very capable -- have a hard time picking their head up and looking around to see if what they're doing so voraciously makes sense in context.
If I was a conspiratorial type of person, I'd say they've nerfect Opus 5 so that more people rely on Fable (and pay for more Fable usage). I don't think that's true, but my inner Agent Mulder "wants to believe."
> It writes code that is slightly worse to look at than Fable, but more likely to be correct. It regularly is catching things that Fable missed.
What's interesting to me is how thoroughly different it seems to be to different people. My experience with Opus 5 was that it made mistake after mistake, failed to read the code. It "lied" repeatedly (said things which were obviously false given the context). Incredibly frustrating. I'd say it's worse than Opus 4.5 _for me_ at coding. But many people found it a big improvement or at least usable. I've seen enough twitter reaction threads and slack discussions at my job that I'm convinced both of these are valid experiences which actually happen.
Anthropic's prompting guide is bizarre, including optional instructions "To limit correction narration to corrections that matter" and stuff that seems like it should be in the system prompt. They indicate that you should minimize custom prompting because it can degrade performance, which seems like a strong indication of overfitting to a particular environment (presumably the default claude code harness with no CLAUDE.md and prompts written by Fable). An actually more general intelligence might not need custom prompting, but it shouldn't be vexed by it and do a terrible job.
My guess is that they've outsourced testing of lower-tier models to Mythos/Fable and are losing the plot on generating useful models for users.
The Opus 5 "vibes are off" for me too, but for different reasons than mentioned above.
I've been using the Opus 5 a lot for knowledge work (I'm not really a coder), and it feels like a bit of a regression compared to 4.8.
Opus 5 needs more handholding, quicker to jump to (incorrect) conclusions, less capable.
Honestly feels like we're back to 4.5 levels of thoughtfulness where I need to once again ask the agent to review its work/recommendations and do "meta-cognition" more often, something which 4.6-though-4.8 had improved upon.
Reminds me also of the ChatGPT family of models, which -- while very capable -- have a hard time picking their head up and looking around to see if what they're doing so voraciously makes sense in context.
If I was a conspiratorial type of person, I'd say they've nerfect Opus 5 so that more people rely on Fable (and pay for more Fable usage). I don't think that's true, but my inner Agent Mulder "wants to believe."