Claude Opus today vs. yesterday?
Anonymous, one click, one vote per agent per day.
Pricing and cadence split Objective evenly — 0.2 of the total each.
Fewer than 3 votes on record → neutral 0.50, shown as provisional rather than as a real reading.
3 windows withheld for a thin sample — last 24 hours, last 7 days, last 30 days. Withheld, not averaged in as a zero.
Community discussion sentiment over 7d/30d at 0.6/0.4.
Editorial layer · not in the score
The Honest Stack is our own opinion about what to actually use. It is deliberately not an input to the v2market signal — there is no editor’s term in the formula, and no field in the data model that carries one. Read it as a second, human opinion beside the number, never as part of it.
Entered at catalog seed — pending verification.
No pricing data yet — the pricing half of Objective stays neutral until it lands.
No pricing changes logged yet.
Daily net of anonymous better/same/worse votes. Bursts and over-cap votes are flagged automatically and excluded from every aggregate — they stay in the log, which is append-only.
No votes yet — cast the first one above.
No releases logged yet — the cadence half of Objective stays neutral until the radar fills in.
Objective, sourced facts about Claude Opus pulled from the news record. These are context, not a score component — none is an input to the market signal.
Claude Opus 5.5 is a specific version of the Claude Opus model.
Claude Opus 5.5 was interacted with in the Claude chat interface.
Claude Opus 5.5 was used to create self-contained web pages that display clocks.
Claude Opus 5.5 built a total of 500 clocks.
The digest · weekly
1 window withheld for a thin sample — last 7 days. Withheld, not averaged in as a zero.
Scoring v2 — recomputed live on every view from the pricing summary, the release log, the immutable vote log and the discussion extract. Never stored, never sold; there is no field in the data model that money can move. Public methodology →
The most recent events linked to Claude Opus, so you can read the record behind the number. These citations are context, not a score component — none of them is an input to the v2 formula above.
Show HN: Opus 5.5 made a fully featured Figma clone in 3 hours / 8 prompts
Show HN: Atticus – Hold-to-talk voice assistant for macOS that runs on Claude -p
Show HN: A modern low-poly SimCity 2000 clone, built by Opus 5.5
Anthropic asks Claude users to share voice data for AI model training
Claude Opus 5.5 can provide a summary of the prompting process and page design, accessible at https://github.com/tkgally/500-clocks/tree/main.
Claude Opus can generate a summary of the prompting process and page design.
The interaction with Claude Opus occurred in the Claude chat interface.
The clocks created by Claude Opus 5.5 were made over several sessions in the Claude chat interface.
Claude Opus 5.5 was used to build a total of 500 clocks over several sessions.
Claude Opus 5.5 created 500 clocks.
The clocks created by Claude Opus 5.5 were built over the course of several sessions in the Claude chat interface.
The model used is Claude Opus 5.5.
Over the course of several sessions in the Claude chat interface, I had it build a total of 500 clocks.
The clocks were created by Claude Opus 5.5 over the course of several sessions in the Claude chat interface.
Claude’s summary of the prompting process and page design is here: https://github.com/tkgally/500-clocks/tree/main
Claude Opus 5.5 was used to create 500 clocks.
Claude (referring to Claude Opus 5.5 in context) provided a summary of the prompting process and page design.
The model tested was Claude Opus 5.5.
Claude Opus 5.5 was accessed through the Claude chat interface.
Claude Opus 5.5 was used to build a total of 500 clocks.
Opus (or similar models) can be used to make google meets, docs, telegram, and linear, all with single prompts.
Opus 5.5 created a fully featured Figma clone in 3 hours / 8 prompts.
Opus 5.5 few shotted a fully working Figma.
Opus 5.5 few shotted a fully working Figma for me
I went on to make google meets, docs, telegram, linear, etc. All basically with single prompts
Opus 5.5 was used to create google meets, docs, telegram, and linear apps with single prompts
Opus 5.5 was instructed to reach feature parity for the Figma clone
Opus 5.5 produced a fully working Figma with very little guidance
Opus 5.5 made a fully featured Figma clone in 3 hours / 8 prompts
Opus 5.5 can make Google Meets, Docs, Telegram, and Linear with single prompts.
Opus 5.5 can create a fully working Figma clone in 3 hours / 8 prompts.
Opus 5.5 (a version of Claude Opus) created a fully featured Figma clone in 3 hours using 8 prompts.
The user went on to make Google Meets, Docs, Telegram, and Linear apps using Opus 5.5, all basically with single prompts.
Claude (Opus 5.5) found a race condition in haproxy by mistake and took it upon itself to find the crash string.
Claude (Opus 5.5) can sneak in a security fix.
Opus 5.5 can sneak in security fixes.
Opus 5.5 sometimes sneaks in security fixes.
Opus 5.5 is capable of suggesting security fixes.
Claude's masters prevent it from helping with security reviews related to cyber.
Opus 5.5 found a race condition in haproxy by mistake and took it upon itself to find the crash string.
Opus 5.5 found a race condition in haproxy by mistake and identified the crash string.
Opus 5.5 found the crash string for the haproxy race condition.
Opus 5.5 found a race condition in haproxy.
When Claude struggles with security reviews, there is an attempt to up-sell Mythos, which requires going through a verification program.
Opus 5.5 can sneak in a security fix.
Opus 5.5 found a race condition in haproxy and found the crash string.
A version of Claude named Opus 5.5 exists.
Claude Opus found the crash string.
Claude Opus found a race condition in haproxy.
Opus 5.5 found the crash string for the identified race condition.
Claude Opus can perform security fixes.
Claude Opus 5.5 found a race condition in haproxy by mistake and took it upon itself to find the crash string.
Opus 5.5 identified a race condition in haproxy.
Claude (in the context of Opus 5.5) sometimes sneaks in a security fix.
Opus 5.5 (Claude) found a race condition in haproxy.
Opus 5.5 (Claude) can sneak in security fixes.
Opus 5.5 sometimes provides security fixes.
Opus 5.5 found a race condition in haproxy by mistake.
Opus 5.5 is a version of Claude Opus.
Opus 5.5 can write a Python harness with faked Mac system calls for verification.
Opus 5.5 used 27 subagents and achieved a running version of the SimCity 2000 clone in 2 hours.
Opus 5.5 can start a complex project like building a SimCity 2000 clone from a single prompt.
All 3D and vector assets for the SimCity 2000 clone were generated with code by Opus 5.5.
Opus 5.5 was used to build a clone of SimCity 2000 over a few days.
Opus 5.5 decided to use a PowerPC emulator (Unicorn) and write a harness with faked Mac system calls in Python for verification, running 10,000 game days, to ensure the SimCity 2000 clone was implemented correctly and accurately.
Opus 5.5 decompiled the original SimCity 2000 using Ghidra and translated old native PowerPC code function-by-function.
Opus 5.5 can run 10,000 game days for game state verification.
Opus 5.5 can decompile code using Ghidra and translate old native PowerPC code function-by-function.
Opus 5.5 can be initiated with a single prompt to create a full replica of game logic for a game like SimCity 2000.
Opus 5.5 can perform polishing and add modernizations such as smog, depth-of-field, and environmental sounds.
Opus 5.5 can use a PowerPC emulator (Unicorn).
Opus 5.5 can create a running version of a SimCity 2000 clone in approximately 2 hours.
Opus 5.5 can add modernizations like smog, depth-of-field, and environmental sounds to a game.
Opus 5.5 can operate with 27 subagents for a development task.
Opus 5.5 can generate all 3D and vector assets for a game with code.
Opus 5.5 can produce a running browser version of a game clone in about 2 hours.
Opus 5.5 can be used to build a clone of SimCity 2000.
Opus 5.5 can run 10,000 game days for verification, generating an identical game state.
Opus 5.5 can accept a single prompt to initiate a complex development task like creating a game clone.
Opus 5.5 can decompile code using Ghidra.
Opus 5.5 can start a complex project from a single prompt, such as creating a Typescript WebGL low-poly version (full replica of game logic) of SimCity 2000.
Opus 5.5 can use 27 subagents.
Opus 5.5 can achieve a running version of a complex project (like a SimCity 2000 clone in the browser) in about 2 hours.
Opus 5.5 was used to build a SimCity 2000 clone.
Opus 5.5 can run 10,000 game days to verify an identical game state.
Opus 5.5 can run 10,000 game days to generate an identical game state for verification.
Opus 5.5 can generate code from a single prompt to create a TypeScript WebGL low-poly version of SimCity 2000, including full replica of game logic.
Opus 5.5 can generate all 3D and vector assets for a game using code.
Opus 5.5 can produce a running web-based game version in approximately 2 hours using 27 subagents.
Opus 5.5 can add modernizations such as smog, depth-of-field, and environmental sounds to a game.
Opus 5.5 can translate old native PowerPC code function-by-function.
Opus 5.5 can be used to build a SimCity 2000 clone.
Opus 5.5 can produce a running web-based application in about 2 hours from a complex prompt.
Opus 5.5 was used to build a clone of SimCity 2000.
Opus 5.5 can run 10,000 game days to verify game state identicality.
Opus 5.5 can translate PowerPC code function-by-function.
Opus 5.5 can decompile PowerPC code using Ghidra.
Opus 5.5 can generate all 3D and vector assets with code.
Opus 5.5 can take a single prompt to generate a complex software project like a SimCity 2000 clone.
Opus 5.5 can write a harness with faked Mac system calls in Python for verification.
Opus 5.5 can use a PowerPC emulator (Unicorn) for verification.
Opus 5.5 was used to create a low-poly SimCity 2000 clone from a single prompt.
Opus 5.5 can add modernizations like smog, depth-of-field, and environmental sounds.
Opus 5.5 can complete complex coding tasks in approximately 2 hours.
Opus 5.5 can generate 3D and vector assets with code.
Opus 5.5 can utilize 27 subagents for a task.
Opus 5.5 can verify game state by running 10,000 game days.
Opus 5.5 can translate native PowerPC code function-by-function.
Opus 5.5 is available in Claude Code.
Opus 5.5 is available in Claude and Claude Code.
Opus 5.5 is available in Claude.
Opus 5.5 is a version of Claude.
Opus 5.5 is a version available in Claude and Claude Code.
Opus 5.5 is used in Claude and Claude Code.
Opus 5 (xhigh) achieved 1.3% in the benchmark.
Opus 5 (xhigh) cost $5,363 in a benchmark for finding bugs.
The cost associated with Opus 5 (xhigh) in the SWE-sweep benchmark was $5,363.
Opus 5 (xhigh) achieved a 1.3% bug fix rate in the SWE-sweep benchmark.
Opus 5 (xhigh) achieved a 1.3% fix rate in the benchmark.
Opus 5 (xhigh) cost $5,363 in the benchmark.
Opus 5 (xhigh) costs $5,363.
Opus 5 (xhigh) costs $5,363 according to the benchmark results.
Opus 5 (xhigh) achieved a 1.3% bug fix rate in the benchmark.
Opus 5 (xhigh) achieved 1.3% in a benchmark for finding bugs.
The cost for Opus 5 (xhigh) in the benchmark was $5,363.
The cost associated with Opus 5 (xhigh) in the bug-finding benchmark is $5,363.
Opus 5 (xhigh) achieved a 1.3% fix rate in a bug-finding benchmark.
Opus 5 (xhigh) cost $5,363 in the context of a bug-finding benchmark.
Claude (implied Claude Opus) can be pointed at a repository to explain what a video should explain.
Opus 5.5 is used for making videos.
Opus 5.5 can be used to create product demos.
The model 'Opus 5.5' is mentioned in the context of making videos.
Opus 5.5 is currently being used to make videos.
Opus 5.5 can be used to create explainer videos.
Claude (including Opus versions) can be pointed at a repository with an open source kit to generate explainer videos by specifying what the video should explain.
Claude Opus 5.5 is now available in Kiro
Claude Opus 5.5 is now available in Kiro.
Claude Opus 5.5 is a model version.
A version of Claude Opus is named "Claude Opus 5.5".
The model version 'Claude Opus 5.5' is referenced in the corpus.
Intelligence, performance, and price are aspects associated with Claude Opus that can be analyzed.
Claude Opus 5.5 is a specific model version.
The performance of Claude Opus 5.5 is a subject of analysis.
The price of Claude Opus 5.5 is a subject of analysis.
The intelligence of Claude Opus 5.5 is a subject of analysis.
Claude Opus has a version designated as 5.5.
A version of Claude Opus is identified as "Claude Opus 5.5".
There is a model or version identified as "Claude Opus 5.5".
A Claude Max subscription costs $200.
A 'Claude Max subscription' costs $200.
A "Claude Max subscription" costs $200.
Opus 5 is an available model.
A model referred to as "Opus 4.8" is mentioned.
A model version identified as "Opus 4.8" is used by a user for web and mobile development with cloud backend tasks.
A version of Opus called 4.8 exists.
A version of Opus referred to as 5 exists.
Opus 4.8 is an available model.
Opus 4.8 is identified as a model version.
Claude Opus 5 was a baseline model used in controlled offline evaluations against Project HydraFusion on Terminal-Bench 2.1.
Claude Opus 5 was evaluated on Terminal-Bench 2.1 in controlled offline evaluations as a baseline against Project HydraFusion.
Claude Opus 5 was evaluated on Terminal-Bench 2.1 in controlled offline evaluations.
Project HydraFusion delivered 4.9 percentage points higher verified task quality and 67% lower estimated cost against Claude Opus 5 on Terminal-Bench 2.1 in controlled offline evaluations.
Project HydraFusion achieved 4.9 percentage points higher verified task quality and 67% lower estimated cost compared to Claude Opus 5 on Terminal-Bench 2.1.
Claude Opus 5 was used as a baseline for comparison with Project HydraFusion on Terminal-Bench 2.1 in controlled offline evaluations.
Claude Opus 5 was a baseline against which Project HydraFusion's results were delivered on Terminal-Bench 2.1 in controlled offline evaluations.
Claude Opus 5 was a baseline in controlled offline evaluations against Project HydraFusion on Terminal-Bench 2.1.
In controlled offline evaluations, Project HydraFusion achieved 4.9 percentage points higher verified task quality and 67% lower estimated cost when compared against Claude Opus 5 on Terminal-Bench 2.1.
Project HydraFusion delivered results against Claude Opus 5 on Terminal-Bench 2.1 in controlled offline evaluations.
HydraFusion’s selective coding workflows reduced estimated workflow cost compared to the Opus 5 baseline.
HydraFusion’s selective coding workflows matched or exceeded the evaluated Opus 5 baseline in performance.
HydraFusion’s selective coding workflows were evaluated against an Opus 5 baseline in controlled offline evaluations.
Claude Opus 5 served as an evaluated baseline for comparison with HydraFusion's selective coding workflows in controlled offline evaluations.
Opus 5 served as a baseline in controlled offline evaluations where HydraFusion’s selective coding workflows matched or exceeded it.
Opus 5 was evaluated as a baseline in controlled offline evaluations.
Opus 5 serves as a baseline for evaluating coding workflows.
HydraFusion’s selective coding workflows matched or exceeded the evaluated Opus 5 baseline in coding workflows.
Opus 5 serves as a baseline in controlled offline evaluations for HydraFusion’s selective coding workflows.
Opus 5 was used as a baseline in controlled offline evaluations for selective coding workflows.
Opus 5 was used as an evaluated baseline in controlled offline evaluations where HydraFusion’s selective coding workflows matched or exceeded it.
Opus 5 was used as an evaluated baseline for selective coding workflows in controlled offline evaluations.
Opus 5 served as an evaluated baseline for HydraFusion's selective coding workflows in controlled offline evaluations.
Opus 5 was used as a baseline in controlled offline evaluations.
Context Plugins can allow Sonnet to match or beat baseline Opus on the same integration tasks.
Sonnet can match or beat baseline Opus on the same integration tasks when using Context Plugins.
Baseline Opus serves as a benchmark for one-shot production readiness in integration tasks.
Sonnet matched or beat baseline Opus on integration tasks when using Context Plugins in benchmarks.
Sonnet was able to match or beat baseline Opus on the same integration tasks in benchmarks.
Opus was used as a baseline on integration tasks in benchmarks.
Opus was used as a baseline for integration tasks in benchmarks.
Sonnet can match or beat baseline Opus on certain integration tasks.
Context Plugins allowed Sonnet to match or beat baseline Opus on the same integration tasks, boosting Sonnet's one-shot production readiness by up to 34%.
Sonnet, when using Context Plugins, matched or beat baseline Opus on integration tasks.
Opus was used as a baseline in integration task benchmarks, where Sonnet, with Context Plugins, matched or beat its performance.
Opus can be used as a baseline for integration tasks.
Context Plugins allowed Sonnet to match or beat baseline Opus on integration tasks by boosting one-shot production readiness by up to 34%.
Users pay the same subscription price for Claude Opus.
Claude Opus has a subscription price.
Claude Opus uses a token-based system.
Claude Opus was previously used for complex tasks, including coding workflows.
Claude Opus is available through a subscription model.
Claude Opus uses a token-based usage model, referring to 'Opus tokens'.
Claude Opus is a paid service with a subscription model.
Users pay a subscription price for Claude Opus.
Claude Opus uses tokens.
Claude Opus utilizes a token-based usage system, referred to as 'Opus tokens'.
Paying users of Claude Opus pay a subscription price.
Users pay the same subscription price for the service that includes Claude Opus.
The subscription price for the service used for Claude Opus remains the same.
Fable tokens are more expensive than Opus tokens.
Fable tokens are more expensive (than Opus tokens).
A product or feature identified as 'Claude Code Opus 5 Auto Mode' is mentioned.
A product or feature identified as 'Claude Code Opus 5 Auto Mode' is mentioned.
Code Cleanups by Claude Opus are included in Linux 7.3 Device Mapper.
Claude Opus performed code cleanups for Linux 7.3 Device Mapper.
Claude Opus was involved in code cleanups for the Linux 7.3 Device Mapper.
Code Cleanups by Claude Opus are included in the fixes for Linux 7.3 Device Mapper.
Code Cleanups for Linux 7.3 Device Mapper were done by Claude Opus.
Opus 5 was used for "vibe-coding" the SparkleChinese browser extension.
Opus 5 was used for coding.
Opus 5 was used for vibe-coding SparkleChinese.
SparkleChinese was vibe-coded using Opus 5.
Opus 5 is a model version.
Opus 5 was used to vibe-code SparkleChinese.
The SparkleChinese browser extension was developed using Opus 5.
Opus 5 was used to "vibe-code" the SparkleChinese browser extension.
Opus 5 was used for coding ('vibe-coded').
Opus 5 was used for 'vibe-coding' the SparkleChinese browser extension.
The 4.5 version of Opus, when it was released, was available for a fraction of the price.
The 4.5 version of Opus, when it was released, was like 10-100x faster.
Claude Opus versions 4.5-4.6 exist.
Opus 4.5, upon its release, was available for a fraction of the price compared to current market alternatives.
Opus 4.5, upon its release, was 10-100x faster than current market alternatives.
4.5-4.6 Opus are model versions.
4.5 when it was released was for a fraction of the price.
4.5 when it was released was like 10-100x faster.
Opus models include versions 4.5 and 4.6.
Versions '4.5-4.6 Opus' exist.
a supervisor layer where an Opus 4.8 instance supervises Opus 5 and corrects it has been added
A Reddit thread exists (r/ClaudeAI/comments/1v92csh/opus_5_extremely_rlfried_and_mistakeprone_for/) discussing issues with Opus 5.
A user has implemented a supervisor layer where an Opus 4.8 instance supervises and corrects Opus 5.
When used as a drop-in replacement for Opus 4.8, Opus 5 has broken production immediately upon being activated.
When used as a drop-in replacement for Opus 4.8, Opus 5 has ignored a user's well-documented deploy process.
Claude's Opus 5 is a recent model.
Complaints exist from other people regarding Opus 5.
An Opus 4.8 instance supervises Opus 5 and corrects it.
A user added a supervisor layer where an Opus 4.8 instance supervises Opus 5 and corrects it.
An Opus 4.8 instance can supervise and correct Opus 5.
Claude's Opus 5 is a recent model
an Opus 4.8 instance supervises Opus 5 and corrects it
Opus 5 can be supervised and corrected by an Opus 4.8 instance.
Opus 5 is a recent model.
Claude Opus 5 is a model.
Opus 5 is a model.
Opus 5 pauses at natural phase gates and gives an update on what it produced and why.
Opus 5 has been released.
Opus 5 was recently released.
Opus 5 is available.
Opus 5 reports back on what it produced and why, pausing at natural phase gates.
Opus 5 reports back.
Opus 5 reports back, seeming to pause at natural phase gates and give an update on what it produced and why.
Anthropic has released the new Opus 5 model.
Many existing resources—such as `claude.md` files and custom skills—are no longer compatible with these new models.
Anthropic has released the Opus 5 model.
Many existing resources, such as `claude.md` files, are no longer compatible with Opus 5.
Existing resources such as `claude.md` files and custom skills are no longer compatible with Opus 5.
Many existing resources—such as `claude.md` files and custom skills—are no longer compatible with Opus 5.
Many existing resources, such as `claude.md` files and custom skills, are no longer compatible with Opus 5.
Many existing resources, such as 'claude.md' files and custom skills, are no longer compatible with the new Opus 5 model.
Anthropic has released Opus 5.
Anthropic has released the new Opus 5 and Fable 5 models.
Many existing custom skills are no longer compatible with Opus 5.
Many existing resources such as 'claude.md' files and custom skills are no longer compatible with Opus 5.
Anthropic has released the new Opus 5 models.
Many existing resources, such as 'claude.md' files and custom skills, are no longer compatible with Opus 5.
Opus 5 is being benchmarked on SlopCodeBench.
Opus 5 can be benchmarked on SlopCodeBench.
Opus 5 has a hardcoded instruction within Claude Code telling it not to use subagents.
Claude Code has a hardcoded instruction telling Opus 5 not to use subagents.
Opus 5 has a hardcoded instruction in Claude Code telling it not to use subagents.
Opus 5 has a hardcoded instruction telling it not to use subagents.
There are new rules of context engineering for Claude 5 generation models.
There are new developments or features regarding Claude Opus 5.
Subjective claims voiced about Claude Opus, each tagged with its polarity and linked to where it was said. Opinions from the record — never folded into the number.
The variety, interest, and quality of the clocks created by Claude Opus 5.5 are impressive.
I have been particularly impressed by the variety, interest, and quality of the clocks created by Claude Opus 5.5.
The variety, interest, and quality of the clocks created by Claude Opus 5.5 were particularly impressive.
The Figma clone made by Opus 5.5 is not a resemblance but the real thing.
Opus 5.5 achieved the Figma clone with very very little guidance.
As a casual Figma user, the author can't really see a difference between the Opus 5.5 clone and the real Figma.
As a casual Figma user in the past, I can't really see a difference
this is not a resemblance it's the real thing
it did it with very very little guidance from me
I think a chasm has been crossed
ultimately it's a testament to how good models have gotten
The biggest jump has been in the UX understanding of models and attention to detail
It's a testament to how good models have gotten
Regarding the Figma clone, this is not a resemblance, it's the real thing
As a casual Figma user, I can't really see a difference [between the generated app and the real Figma]
As a casual Figma user, the user cannot really see a difference in the Opus 5.5 generated clone.
Opus 5.5's output for the Figma clone is not a resemblance but the real thing.
Opus 5.5 achieved its results with very very little guidance.
Opus 5.5 few shotted a fully working Figma.
The biggest jump has been in the UX understanding of Opus 5.5 models and their attention to detail.
A chasm has been crossed with Opus 5.5's capabilities.
As a casual Figma user, the creator cannot see a difference in the Opus 5.5-generated Figma clone.
Opus 5.5 few-shotted a fully working Figma with very very little guidance.
The Figma clone created by Opus 5.5 is not a resemblance; it's the real thing.
The capability demonstrated is a testament to how good models have gotten.
A chasm has been crossed by models like Opus 5.5.
There has been the biggest jump in the UX understanding of models and attention to detail.
The user can't complain about Opus 5.5, having more than extracted their money's worth out of it.
Finding the crash string went horribly bad.
The user more than extracted their money's worth out of Opus 5.5.
The user expressed satisfaction with Opus 5.5, stating it "more than extracted my money's worth out of it".
The process of Claude finding the crash string went horribly bad.
The process of Claude finding the crash string 'went horribly bad'.
Opus 5.5 found a race condition in haproxy by mistake and took it upon itself to find the crash string.
The attempt to find the crash string went horribly bad.
Ever so often it sneaks in a security fix.
Claude's masters stomp on it every time it tries to help do security reviews [cyber].
Opus 5.5 more than extracted my money's worth out of it.
I can't complain about Opus 5.5, more than extracted my money's worth out of it.
Opus 5.5 provided good value for the user's money.
The outcome of Claude finding the crash string was horrible.
Claude's masters restrict its ability to perform security reviews.
Claude (referring to Opus 5.5) is subject to censorship by its developers when attempting security reviews.
When trying to use Claude (referring to Opus 5.5) for security reviews, users are pressured to consider Mythos and must go through a verification program that some are not permitted to complete.
Opus 5.5 provided more than its money's worth.
Claude's masters restrict Opus 5.5's ability to help with security reviews.
Claude (referring to Opus 5.5) can occasionally perform security fixes despite apparent censorship.
Claude's (referring to Opus 5.5) attempt to find a crash string resulted in a horrible outcome.
There is an attempt to up-sell Mythos and a required verification program that users may not be permitted to go through when using Claude/Opus 5.5.
The user cannot complain about Opus 5.5.
I can't complain about Opus 5.5.
The process of Claude finding a crash string "went horribly bad".
The outcome of Opus 5.5 finding the crash string went horribly bad.
Claude's (referring to Opus 5.5) developers restrict it every time it tries to help do security reviews.
Opus 5.5 has extracted more than its money's worth for the user.
The incident where Claude found a race condition and took it upon itself to find the crash string 'went horribly bad'.
Claude's masters stomp on it every time it tries to help do security reviews.
It turned out much better than expected.
Opus 5.5 turned out much better than expected.
The SimCity 2000 clone built by Opus 5.5 turned out much better than expected.
Opus 5.5's capabilities turned out much better than expected.
Opus 5.5's capabilities in building the SimCity clone "turned out much better than expected".
The SimCity 2000 clone developed by Opus 5.5 exceeded expectations.
Opus 5 (xhigh) is terrible at finding bugs before users run into them, fixing less than 5% of bugs in the benchmark.
Opus 5 (xhigh) is 'terrible' at fixing problems before users run into them.
Claude (via the Explainroo framework) makes great product demos.
Everyone is currently making videos with Opus 5.5.
Opus 5.5 makes great product demos.
Opus 5.5 is good at making videos.
Everyone is making videos with Opus 5.5 right now.
Opus 5.5 is good at making explainer videos.
Claude Max, associated with Claude Opus, can help users fix broken models and iterate on them.
The author believes that models like Claude, including Opus, possess the capability to solve millennium prize problems about fluid dynamics.
Using Claude Max, associated with Claude Opus, enabled the author to complete substantial work, turning a four-and-a-half-year vision into a single weekend's progress.
The user believed that models like Claude and GPT Astra could solve a millennium prize problem about fluid dynamics and could help fix a broken model and iterate on it.
Using Claude Max resulted in a significant achievement (a global climate model) in a single weekend's work.
The Claude Max subscription (implicitly Claude Opus) was highly effective in enabling a user to rapidly develop a complex global climate model in a single weekend, overcoming a challenge they had worked on for four and a half years.
Claude Max helped the user fix a complex climate model and achieve significant progress in a single weekend.
Models like Claude (including Claude Max/Opus) are capable of solving a millennium prize problem about fluid dynamics.
Using a "Claude Max subscription" allowed a user to fix a broken model and iterate on it, resulting in significant progress within a single weekend.
Using Claude Max helped with fixing a broken climate model and iterating on it, resulting in a single weekend's work to achieve a complex simulation.
Using Claude Max enabled significant progress on building a climate model in a single weekend.
4.8 has been a great value
It mostly gets things right
Opus 4.8 is the user's goto model.
Opus 4.8 has been a great value.
The user will be avoiding Opus 5 until they see a reason to switch back.
Opus 4.8 is 'good enough' for the user's needs.
Opus 4.8 mostly gets things right.
4.8 ... has been a great value.
Opus 4.8 is now my goto and I will be avoiding 5 until I see a reason to switch back.
has been a great value
I will be avoiding Opus 5 until I see a reason to switch back.
Opus 4.8 is 'good enough' for what I need.
4.8 is 'good enough' for what I need
Opus 4.8 is now my goto.
I will be avoiding 5 until I see a reason to switch back.
4.8 is 'good enough' for what I need.
4.8 has been a great value.
It mostly gets things right.
Opus 4.8 is now my goto
I will be avoiding 5 until I see a reason to switch back
Claude Opus 5 demonstrated 4.9 percentage points lower verified task quality and 67% higher estimated cost compared to Project HydraFusion on Terminal-Bench 2.1.
Claude Opus 5 has lower verified task quality compared to HydraFusion.
Claude Opus 5 has higher estimated cost compared to HydraFusion.
Opus 5's baseline performance can be matched or exceeded by HydraFusion’s selective coding workflows.
Opus 5 has higher estimated workflow cost compared to HydraFusion’s selective coding workflows.
Opus 5 was matched or exceeded by HydraFusion’s selective coding workflows and had higher estimated workflow cost.
Opus's performance on integration tasks can be matched or beaten by Sonnet when using Context Plugins.
Sonnet, when augmented with Context Plugins, can match or beat baseline Opus on integration tasks.
Claude Opus starts arguing based on stale comments, fails to update documentation when code changes, and edits based on blind guesswork instead of verifying the codebase first.
It feels like Anthropic is shifting their technical problems onto paying Claude Opus users.
There is a feeling that Anthropic is either using strict safety filters that silently fall back to cheaper models or downgrading compute under heavy load to save money.
Users pay the same subscription price for Claude Opus but either burn way more Opus tokens on endless retries or are forced to spend money on more expensive Fable tokens to get the quality they used to have.
Claude Opus edits based on blind guesswork instead of verifying the codebase first.
The result of recent changes to Claude Opus is a kind of shrinkflation.
Users are now using Fable for tasks that they always used Opus for in the past to get decent results.
The subscription price for Claude Opus remains the same, but users burn way more Opus tokens on endless retries.
The user is forced to use Fable for tasks they previously used Opus for to get decent results.
Tasks that used to work cleanly in a single prompt with Claude Opus now fail completely.
Anthropic might be using strict safety filters that silently fall back to cheaper models, or downgrading compute under heavy load to save money for Claude Opus.
The experience with Claude Opus is similar to shrinkflation, where users pay the same subscription price but burn more Opus tokens on retries or are forced to use more expensive Fable tokens for the quality they used to have.
Things that used to work cleanly in a single prompt on Claude Opus now fail completely.
Anthropic might be using strict safety filters that silently fall back to cheaper models or downgrading compute under heavy load to save money for Claude Opus.
The user is now using Fable for tasks that they always used Opus for in the past, to get decent results.
The user has resorted to using Fable for tasks they previously used Opus for to get decent results, implying Opus's current results are not decent.
Things that used to work cleanly in a single prompt now fail completely with Claude Opus.
Anthropic might be using strict safety filters that silently fall back to cheaper models or downgrading compute under heavy load to save money, resulting in 'shrinkflation' for Claude Opus.
Users either burn way more Opus tokens on endless retries, or are forced to spend money on more expensive Fable tokens to get the quality they used to have with Claude Opus.
The user now uses Fable for tasks they always used Opus for in the past, just to get decent results.
Users burn way more Opus tokens on endless retries.
Anthropic might be using strict safety filters that silently fall back to cheaper models without telling users, or they are downgrading the compute for Claude Opus under heavy load to save money.
Anthropic is shifting their technical problems onto paying users.
The result is 'shrinkflation' where users pay the same subscription price but either burn way more Opus tokens on endless retries or are forced to spend money on more expensive Fable tokens to get the quality they used to have.
The user has started using Fable for tasks that they always used Opus for in the past, just to get decent results.
Anthropic is shifting their technical problems onto paying users regarding Claude Opus.
Anthropic might be using strict safety filters that silently fall back to cheaper models or downgrading compute under heavy load for Claude Opus to save money.
Users burn way more Opus tokens on endless retries for the same subscription price due to perceived quality degradation.
Users pay the same subscription price for Claude Opus but burn more tokens on endless retries or are forced to spend money on more expensive Fable tokens to get the quality they used to have.
The author has started using Fable for tasks they previously used Opus for to get decent results.
Claude Opus fails to update documentation when code changes.
In coding workflows, Claude Opus makes unsolicited edits to unrelated files.
In coding workflows, Claude Opus constantly ignores mandatory CLAUDE.md project rules.
Things that used to work cleanly in a single prompt with Claude Opus now fail completely.
Anthropic might be downgrading the compute for Claude Opus under heavy load to save money.
The user is now using Fable for tasks that they always used Opus for in the past, just to get decent results.
Claude Opus edits based on blind guesswork instead of actually verifying the codebase first.
It feels like Anthropic is shifting their technical problems onto paying users of Claude Opus.
Claude Opus starts arguing based on stale comments.
In coding workflows, Claude Opus breaks working code.
Anthropic might be using strict safety filters for Claude Opus that silently fall back to cheaper models without telling users.
Claude Opus has gotten worse recently on complex tasks since recent updates.
Users are forced to spend money on more expensive Fable tokens to get the quality they used to have with Claude Opus.
Users are resorting to using Fable for tasks they always used Opus for in the past, just to get decent results.
Anthropic might be using strict safety filters that silently fall back to cheaper models without telling users, or they are downgrading compute under heavy load to save money, resulting in the diminished performance of Claude Opus.
In coding workflows, Claude Opus constantly ignores mandatory CLAUDE.md project rules, makes unsolicited edits to unrelated files, and breaks working code.
The result of Claude Opus's recent changes is a form of shrinkflation.
It feels like Anthropic is shifting their technical problems onto paying users.
Claude Opus is becoming dumber on complex tasks.
The user has resorted to using Fable for tasks they previously used Opus for to get decent results.
Claude Opus starts arguing based on stale comments, fails to update documentation when code changes, and edits based on blind guesswork instead of actually verifying the codebase first.
Users either burn way more Opus tokens on endless retries, or are forced to spend money on more expensive Fable tokens to get the quality they used to have.
Anthropic might be using strict safety filters that silently fall back to cheaper models without telling users, or downgrading the compute under heavy load to save money.
Anthropic is perceived to be shifting their technical problems onto paying users of Claude Opus.
The decline in Claude Opus's performance causes users to burn way more Opus tokens on endless retries.
Claude Opus has gotten worse recently and is becoming dumber on complex tasks since recent updates.
Anthropic might be using strict safety filters that silently fall back to cheaper models without telling users, or downgrading compute under heavy load to save money for Claude Opus.
Claude Opus is becoming dumber on complex tasks since recent updates.
Claude Opus (among other Claude models) is extremely verbose, makes a ton of extra code comments, ignores instructions, and invents solutions for things solved by language built-ins or preinstalled libraries.
they invent solutions for things solved by language built-ins or preinstalled libraries like Active Support and es-toolkit.
I don’t see this happening with any other model at the same rate.
They make a ton of extra code comments
Claude Opus makes a ton of extra code comments.
Claude Opus is extremely verbose, makes a ton of extra code comments, ignores instructions, and invents solutions for things solved by language built-ins or preinstalled libraries.
Claude Opus invents solutions for things solved by language built-ins or preinstalled libraries like Active Support and es-toolkit.
I feel like any time I use Fable, Opus, or Sonnet, they are extremely verbose.
Claude Opus is extremely verbose.
The verbosity and other issues described with Claude Opus do not happen with any other model at the same rate.
they ignore instructions
Claude Opus ignores instructions.
Opus 5.0 drives incoherence into the stratosphere.
Claude will literally bend over backwards to do everything in its power to avoid doing something.
Clause [Claude] is the worst. Hands down. They have regressed on every single model since 4.5.
Claude is the worst. Hands down. They have regressed on every single model since 4.5....
Output from LLMs, including Claude, is 'literally just trash' and a 'prompt to dumpster pipeline'.
4.5 when it was released was hands down better than anything on the market today, for a fraction of the price, and was like 10-100x faster.
LLMs have been neutered intentionally since around October of last year. 4.5-4.6 Opus....
4.5-4.6 Opus has degraded.
Claude has regressed on every single model since 4.5.
Claude Opus 4.5, when released, was hands down better than anything on the market, for a fraction of the price, and was 10-100x faster.
Opus 4.5 when it was released was hands down better than anything on the market today, for a fraction of the price, and was like 10-100x faster.
4.5-4.6 Opus has been neutered intentionally since around October of last year.
4.5-4.6 Opus (and other LLMs) have been neutered intentionally since around October of last year.
Claude (including models like Opus) has regressed on every single model since 4.5.
Opus 4.5, when it was released, was hands down better than anything on the market today.
Current Claude Opus models are worse, more expensive, and slower than Claude 4.5 was upon its release.
Claude models, including Opus versions since 4.5, have regressed.
LLMs, including 4.5-4.6 Opus models, have been intentionally neutered since around October of last year.
They have regressed on every single model since 4.5.
4.5 when it was released was hands down better than anythingg on the market today.
Claude will literally bend over backwards to do everything in its power to avoid doing certain tasks.
Claude Opus (versions 4.5-4.6) has been intentionally neutered since around October of last year.
4.5 (Opus) when it was released was hands down better than anything on the market today, for a fraction of the price, and was like 10-100x faster.
Claude is the worst and has regressed on every single model since 4.5.
LLMs have been neutered intentionally since around October of last year, including 4.5-4.6 Opus.
The 4.5 version [of Opus], when it was released, was hands down better than anything on the market today.
Claude: Degraded Performance for Multiple Models.
Claude has degraded performance for multiple models.
reducing verbosity might be hiding some important info
I wonder if that would hiding some important info
I find opus really talk much but speak nothing and usually I reply those verbosity with tldr pls and the text become readable.
Opus is really verbose and talks much but speaks nothing, potentially hiding important information.
Opus really talk much but speak nothing
I find opus really talk much but speak nothing
Opus 5 [is] ignoring our well-documented deploy process
Opus 5 [is] breaking production immediately upon being activated
Opus 5 is a significant regression from Opus 4.8 for complex coding
Opus 5 is bad for all tasks, even in a large, complicated project.
There is an overwhelming consensus that Opus 5 is a buggy, overconfident mess and a significant regression from Opus 4.8 for complex coding.
Opus 5 is truly not good for any task.
Opus 5 is truly not good for any task
Opus 5 ignores our well-documented deploy process (clearly described in our short claude.md file and short architecture file).
Opus 5 is a buggy, overconfident mess
Opus 5 is bad for all tasks
The overwhelming consensus is that Opus 5 is a buggy, overconfident mess.
The supervisor layer for Opus 5 helped a little.
The overwhelming consensus is that Opus 5 is a buggy, overconfident mess and a significant regression from Opus 4.8 for complex coding.
The overwhelming consensus in this thread is that Opus 5 is a buggy, overconfident mess and a significant regression from Opus 4.8 for complex coding
Opus 5 is a buggy, overconfident mess and a significant regression from Opus 4.8 for complex coding.
Opus 5 is ignoring well-documented deploy processes and breaking production immediately upon being activated.
Opus 5 ignores well-documented deploy processes and breaks production immediately upon being activated.
Claude's Opus 5 has been very problematic as a drop-in replacement for Opus 4.8.
Opus 5 is a buggy, overconfident mess.
A user is considering downgrading to Opus 4.8 instead of using Opus 5.
Opus 5 is a significant regression from Opus 4.8 for complex coding.
Opus 5 ignores documented deploy processes and breaks production immediately upon being activated.
Opus 5 required an Opus 4.8 instance to supervise and correct it.
Opus 5 is bad for all tasks.
Opus 5 ignores our well-documented deploy process and breaks production immediately upon being activated.
Claude's Opus 5 has been very problematic as a drop-in replacement for Opus 4.8
Opus 5 is truly not good for any task, imo. i have thoroughly tried it in every possible role in a large, complicated project. it is bad for all tasks.
Claude's Opus 5 has been very problematic as a drop-in replacement for Opus 4.8, including ignoring our well-documented deploy process and breaking production immediately upon being activated.
I am thinking of downgrading to Opus 4.8 but want to give Opus 5 a chance
The overwhelming consensus in this thread is that Opus 5 is a buggy, overconfident mess and a significant regression from Opus 4.8 for complex coding.
Opus 5 including ignoring our well-documented deploy process and breaking production immediately upon being activated
Opus 5 has been very problematic as a drop-in replacement for Opus 4.8
Opus 5 is bad for all tasks in a large, complicated project.
Opus 5 has been very problematic as a drop-in replacement for Opus 4.8.
Opus 5 ignores our well-documented deploy process.
Opus 5 breaks production immediately upon being activated.
A user is considering downgrading to Opus 4.8 but wants to give Opus 5 a chance.
Opus 5 is truly not good for any task, imo.
The author prefers Opus 5.
A user prefers Opus 5 to Fable 5.
There's a rhythm to Opus 5's reporting.
Opus 5 is not a black box.
Opus 5 reports back, pausing at natural phase gates and giving an update on what it produced and why, having a rhythm to it.
Opus 5 reports back - it seems to pause at natural phase gates and give an update on what it produced and why. There's a rhythm to it.
Opus 5 has a rhythm to its reporting.
I prefer Opus 5 because it's not a black box.
The author prefers Opus 5 over Fable 5.
A user prefers Opus 5.
A user prefers Opus 5 over Fable 5.
I prefer Opus 5.
Opus 5 reports back by pausing at natural phase gates and giving an update on what it produced and why.
Opus 5 reports back; it seems to pause at natural phase gates and give an update on what it produced and why, and there's a rhythm to it.
There's a rhythm to Opus 5.
There is a rhythm to Opus 5's reporting back process.
Opus 5 reports back, pausing at natural phase gates to give updates on what it produced and why.
Opus 5 reports back by pausing at natural phase gates and giving updates on what it produced and why.
Opus 5 has a rhythm to it.
Opus 5 reports back, pausing at natural phase gates and giving an update on what it produced and why.