[{"data":1,"prerenderedAt":1091},["ShallowReactive",2],{"blog-post-is-opus-5-5-nerfed":3,"related-posts-is-opus-5-5-nerfed":1067},{"id":4,"title":5,"author":6,"body":11,"category":1043,"date":1044,"description":1045,"extension":1046,"featured":1047,"hideToc":1047,"image":1048,"imageAlt":1049,"imageHeight":1050,"imageWidth":1051,"itemList":1052,"lang":1052,"meta":1053,"navigation":548,"noindex":1047,"path":1054,"readingTime":1055,"redirected":1047,"relatedSlugs":1052,"seo":1056,"seoTitle":1057,"stem":1058,"subtitle":1052,"tags":1059,"updatedDate":1044,"__hash__":1066},"blog/blog/is-opus-5-5-nerfed.md","Is Opus 5.5 Nerfed? How to Check Which Model Your Claude Code Session Actually Ran",{"name":7,"role":8,"avatar":9,"bio":10},"Shabnam Katoch","Growth Head","/img/avatars/shabnam-profile.jpeg","Shabnam leads growth at BetterClaw and has spent the past three years building organic distribution for AI-native products. She writes about no-code agent builders, content strategy, and what it actually takes to grow without an ads budget.",{"type":12,"value":13,"toc":1023},"minimark",[14,18,34,138,141,147,150,153,158,167,170,173,179,183,186,242,253,260,263,270,283,290,327,339,343,346,354,357,363,374,377,380,394,397,425,429,446,452,460,466,649,655,661,665,672,681,684,687,695,699,702,725,739,745,751,768,774,778,781,784,793,796,800,803,854,864,870,874,880,883,890,894,897,900,915,919,923,937,941,958,962,976,980,995,999,1002,1006,1019],[15,16,17],"p",{},"Before you blame the model, check which model answered. Two commands and one log line settle most of it.",[19,20,22],"callout",{"type":21},"quick-fix",[15,23,24,28,29,33],{},[25,26,27],"strong",{},"Is Opus 5.5 nerfed?"," No public evidence shows Anthropic changed the model, and Anthropic has made no statement about it as of 10 November 2026. What you can verify: a safeguards flag can move your session to Opus 4.8 or Opus 5, and ",[30,31,32],"code",{},"/cost"," shows which model answered, so check both before you draw conclusions.",[35,36,37,53],"table",{},[38,39,40],"thead",{},[41,42,43,47,50],"tr",{},[44,45,46],"th",{},"Check",[44,48,49],{},"Where",[44,51,52],{},"What it tells you",[54,55,56,74,87,103,120],"tbody",{},[41,57,58,62,71],{},[59,60,61],"td",{},"Which models answered",[59,63,64,66,67,70],{},[30,65,32],{}," (alias of ",[30,68,69],{},"/usage","), \"Usage by model\" lines",[59,72,73],{},"Every model that billed tokens this session",[41,75,76,79,84],{},[59,77,78],{},"Current model and version",[59,80,81],{},[30,82,83],{},"/status",[59,85,86],{},"The model your next message goes to",[41,88,89,92,97],{},[59,90,91],{},"Effort level",[59,93,94],{},[30,95,96],{},"/effort status",[59,98,99,100],{},"Opus 5.5 defaults to ",[30,101,102],{},"medium",[41,104,105,108,117],{},[59,106,107],{},"Model per reply",[59,109,110,113,114],{},[30,111,112],{},"message.model"," in ",[30,115,116],{},"~/.claude/projects/\u003Cproject>/\u003Csession-id>.jsonl",[59,118,119],{},"Which model wrote each reply (internal format)",[41,121,122,125,135],{},[59,123,124],{},"Safeguards flag",[59,126,127,128,131,132],{},"A line naming ",[30,129,130],{},"safeguards"," with ",[30,133,134],{},"Details: [cyber]",[59,136,137],{},"A classifier flagged the request",[15,139,140],{},"Picture this: it's Thursday afternoon. The refactor Opus 5.5 handled cleanly last month now comes back half done, twice, and your usage bar is climbing faster than it should.",[15,142,143],{},[144,145,146],"em",{},"They nerfed it. They must have.",[15,148,149],{},"Maybe. Or your session quietly moved to Opus 4.8 an hour ago, compacted while you were in a meeting, or runs at an effort level you never chose. Each leaves a trace you can read in five minutes.",[15,151,152],{},"We read the Claude Code docs, the changelog, Anthropic's help centre, its status page and the open GitHub issues on 10 November 2026. Here is what is documented and how to check your own sessions.",[154,155,157],"h2",{"id":156},"is-opus-55-nerfed-or-is-your-session-running-something-else","Is Opus 5.5 nerfed, or is your session running something else?",[15,159,160,161,166],{},"Since early November, a wave of Claude Code users has reported the same cluster of symptoms. Answers feel shallower, multi-step tasks stall, and sessions burn through plan usage faster. If you're hitting walls sooner, our guide to the ",[162,163,165],"a",{"href":164},"/blog/claude-rate-limit-reached","Claude rate limit reached error"," decodes each limit message.",[15,168,169],{},"Some also mention refusals on ordinary coding work. GitHub issue #108497 on anthropics/claude-code, opened 9 November, describes an Opus 5.5 session on Windows where a cybersecurity flag stopped every turn on work the reporter says had nothing to do with security.",[15,171,172],{},"Here's the thing. \"It feels worse\" can come from at least five places, and only one is the model itself. The rest are documented things Claude Code does around the model.",[174,175,176],"blockquote",{},[15,177,178],{},"A nerf is a claim about the model. Most of what people feel is a claim about the session. Check the session first.",[154,180,182],{"id":181},"the-safeguards-fallback-when-opus-55-hands-your-work-to-another-model","The safeguards fallback: when Opus 5.5 hands your work to another model",[15,184,185],{},"Opus 5.5 runs with safety classifiers, most often triggered by cybersecurity and biology content. Anthropic's Opus 5.5 announcement of 22 September said \"most cybersecurity tasks will be re-routed to Opus 4.8.\" The Claude Code model configuration docs spell out the targets:",[35,187,188,201],{},[38,189,190],{},[41,191,192,195,198],{},[44,193,194],{},"Model that refused",[44,196,197],{},"Cyber flag re-runs on",[44,199,200],{},"Biology flag re-runs on",[54,202,203,214,223,234],{},[41,204,205,208,211],{},[59,206,207],{},"Opus 5.5",[59,209,210],{},"Opus 4.8",[59,212,213],{},"Opus 5",[41,215,216,219,221],{},[59,217,218],{},"Fable models",[59,220,210],{},[59,222,213],{},[41,224,225,228,231],{},[59,226,227],{},"Sonnet 5.5",[59,229,230],{},"Sonnet 5",[59,232,233],{},"Refusal, no fallback",[41,235,236,238,240],{},[59,237,213],{},[59,239,210],{},[59,241,233],{},[15,243,244,245,248,249,252],{},"The part that matters for \"it got dumber\": after a fallback, ",[25,246,247],{},"the session continues on the fallback model",". Anthropic's help centre article on why Claude switched models says the picker \"stays on the less capable model for the rest of the conversation.\" Run ",[30,250,251],{},"/model"," to go back.",[15,254,255],{},[256,257],"img",{"alt":258,"src":259},"Is Opus 5.5 nerfed diagram: request, safeguards check, flag, re-run on Opus 4.8 or Opus 5, session stays on the fallback","/img/blog/is-opus-5-5-nerfed-fallback-path.jpg",[15,261,262],{},"Two more documented details explain a lot of \"it was fine yesterday\" reports.",[15,264,265,266,269],{},"First, the flag can fire on the very first request, before you type anything unusual. That request carries your CLAUDE.md and git status, so a repo with security or biology material can trip the classifier on context alone. Start a session with ",[30,267,268],{},"claude --safe-mode"," to test whether a customization is the trigger.",[15,271,272,273,275,276,278,279,282],{},"Second, the fallback keeps the effort level of the flagged request. A session on Opus 5.5 at its default ",[30,274,102],{}," that falls back to Opus 4.8 stays at ",[30,277,102],{},", although Opus 4.8 defaults to ",[30,280,281],{},"high",". So you can end up on an older model at a lower effort than it would normally run.",[15,284,285,286,289],{},"If you'd rather decide each time, run ",[30,287,288],{},"/config",", open \"Switch models when a message is flagged\" and choose \"Ask each time\", or set this in your settings file:",[291,292,297],"pre",{"className":293,"code":294,"language":295,"meta":296,"style":296},"language-json shiki shiki-themes github-light","{\n  \"switchModelsOnFlag\": false\n}\n","json","",[30,298,299,308,321],{"__ignoreMap":296},[300,301,304],"span",{"class":302,"line":303},"line",1,[300,305,307],{"class":306},"sgsFI","{\n",[300,309,311,315,318],{"class":302,"line":310},2,[300,312,314],{"class":313},"sYu0t","  \"switchModelsOnFlag\"",[300,316,317],{"class":306},": ",[300,319,320],{"class":313},"false\n",[300,322,324],{"class":302,"line":323},3,[300,325,326],{"class":306},"}\n",[15,328,329,330,333,334,338],{},"To turn the automatic switch off entirely, set ",[30,331,332],{},"CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1",". For agents that route around these flags, our breakdown of ",[162,335,337],{"href":336},"/blog/claude-opus-5-5-agents-cost-rerouting","Opus 5.5 cost and the rerouting problem"," covers what a silent switch does to an unattended workflow.",[154,340,342],{"id":341},"what-safeguards-stopped-the-response-above-means","What \"safeguards stopped the response above\" means",[15,344,345],{},"The log line people paste looks like this, character for character, from GitHub issue #108497:",[291,347,352],{"className":348,"code":350,"language":351,"meta":296},[349],"language-text","Opus 5.5's safeguards stopped the response above · continuing once with that noted\n","text",[30,353,350],{"__ignoreMap":296},[15,355,356],{},"It starts with the name of the model that flagged the request. On Opus 5.5 it was followed by an error that names the category:",[291,358,361],{"className":359,"code":360,"language":351,"meta":296},[349],"API Error: Opus 5.5's safeguards flagged this session (https://www.anthropic.com/legal/aup).\n[...] Claude Code can't respond to your last message with Opus 5.5.\n\nDouble press esc to edit your last message, or try a different model with /model.\n\nDetails: `[cyber]`\n",[30,362,360],{"__ignoreMap":296},[15,364,365,366,369,370,373],{},"Read literally, the first line says the model's safeguards cut off a response and Claude Code is trying once more. The ",[30,367,368],{},"Details"," tag names the classifier that fired. The Claude Code errors page confirms that on Opus 5.5 and Sonnet 5.5 this family of messages opens with ",[30,371,372],{},"\u003Cmodel>'s safeguards flagged this session",".",[15,375,376],{},"Here's the twist. In #108497 the session did not fall back: the error came back on seven turns in a row, and the reporter stayed on Opus 5.5.",[15,378,379],{},"Issue #108506, filed by the same reporter about an hour later, shows the same errors, starting on Sonnet 5.5. Both were open with no Anthropic reply when we checked.",[15,381,382,383,386,387,390,391,373],{},"Claude Code 2.1.321, released 9 November, lists a fix for \"the fallback model sometimes declining a request or mentioning a stopped response when the first model was blocked before replying.\" That may be related, but Anthropic hasn't linked it to either issue. ",[25,384,385],{},"Update first, then judge."," Run ",[30,388,389],{},"claude update"," and check the version with ",[30,392,393],{},"claude --version",[15,395,396],{},"What to do when you see the line:",[398,399,400,408,418],"ol",{},[401,402,403,404,407],"li",{},"Press Esc twice or run ",[30,405,406],{},"/rewind"," to step back before the flagged turn, then rephrase or split the sensitive step out.",[401,409,410,411,414,415,373],{},"If you can't find the trigger, run ",[30,412,413],{},"/clear"," for a fresh conversation. The old one stays in ",[30,416,417],{},"/resume",[401,419,420,421,424],{},"If the work wasn't about security, run ",[30,422,423],{},"/feedback"," to report the false positive.",[154,426,428],{"id":427},"how-to-check-which-model-ran-in-your-claude-code-session","How to check which model ran in your Claude Code session",[15,430,431,432,434,435,437,438,441,442,445],{},"Run ",[30,433,32],{},". It's an alias for ",[30,436,69],{},", and the Session block lists usage by model, one line per model. If ",[30,439,440],{},"claude-opus-4-8"," or ",[30,443,444],{},"claude-opus-5"," shows up in a session you thought was all Opus 5.5, you've found your fallback.",[15,447,448,449,451],{},"Claude Code 2.1.285 fixed a bug where ",[30,450,32],{}," could file a turn under the wrong model after a refusal, so trust it only on a current version.",[15,453,431,454,456,457,459],{},[30,455,83],{}," to see the model your next message goes to, and ",[30,458,96],{}," for the effort level.",[15,461,462,463,465],{},"For a reply-by-reply view, read the transcript. Claude Code stores each session as JSONL at ",[30,464,116],{},", and each assistant entry carries the model that wrote it. These commands read it (checked against Claude Code 2.1.322 on 10 November):",[291,467,471],{"className":468,"code":469,"language":470,"meta":296,"style":296},"language-bash shiki shiki-themes github-light","# Count replies per model in one session\njq -r 'select(.type==\"assistant\") | .message.model' \\\n  ~/.claude/projects/\u003Cproject>/\u003Csession-id>.jsonl | sort | uniq -c\n\n# See when the model changed\njq -r 'select(.type==\"assistant\") | [.timestamp, .message.model] | @tsv' \\\n  ~/.claude/projects/\u003Cproject>/\u003Csession-id>.jsonl | uniq -f1\n\n# See when the session compacted\njq -r 'select(.type==\"system\" and .subtype==\"compact_boundary\") | .timestamp' \\\n  ~/.claude/projects/\u003Cproject>/\u003Csession-id>.jsonl\n","bash",[30,472,473,479,495,543,550,556,568,600,605,611,623],{"__ignoreMap":296},[300,474,475],{"class":302,"line":303},[300,476,478],{"class":477},"sAwPA","# Count replies per model in one session\n",[300,480,481,485,488,492],{"class":302,"line":310},[300,482,484],{"class":483},"s7eDp","jq",[300,486,487],{"class":313}," -r",[300,489,491],{"class":490},"sYBdl"," 'select(.type==\"assistant\") | .message.model'",[300,493,494],{"class":313}," \\\n",[300,496,497,500,504,507,510,513,516,518,521,524,526,529,532,535,537,540],{"class":302,"line":323},[300,498,499],{"class":490},"  ~/.claude/projects/",[300,501,503],{"class":502},"sD7c4","\u003C",[300,505,506],{"class":490},"projec",[300,508,509],{"class":306},"t",[300,511,512],{"class":502},">",[300,514,515],{"class":490},"/",[300,517,503],{"class":502},[300,519,520],{"class":490},"session-i",[300,522,523],{"class":306},"d",[300,525,512],{"class":502},[300,527,528],{"class":490},".jsonl",[300,530,531],{"class":502}," |",[300,533,534],{"class":483}," sort",[300,536,531],{"class":502},[300,538,539],{"class":483}," uniq",[300,541,542],{"class":313}," -c\n",[300,544,546],{"class":302,"line":545},4,[300,547,549],{"emptyLinePlaceholder":548},true,"\n",[300,551,553],{"class":302,"line":552},5,[300,554,555],{"class":477},"# See when the model changed\n",[300,557,559,561,563,566],{"class":302,"line":558},6,[300,560,484],{"class":483},[300,562,487],{"class":313},[300,564,565],{"class":490}," 'select(.type==\"assistant\") | [.timestamp, .message.model] | @tsv'",[300,567,494],{"class":313},[300,569,571,573,575,577,579,581,583,585,587,589,591,593,595,597],{"class":302,"line":570},7,[300,572,499],{"class":490},[300,574,503],{"class":502},[300,576,506],{"class":490},[300,578,509],{"class":306},[300,580,512],{"class":502},[300,582,515],{"class":490},[300,584,503],{"class":502},[300,586,520],{"class":490},[300,588,523],{"class":306},[300,590,512],{"class":502},[300,592,528],{"class":490},[300,594,531],{"class":502},[300,596,539],{"class":483},[300,598,599],{"class":313}," -f1\n",[300,601,603],{"class":302,"line":602},8,[300,604,549],{"emptyLinePlaceholder":548},[300,606,608],{"class":302,"line":607},9,[300,609,610],{"class":477},"# See when the session compacted\n",[300,612,614,616,618,621],{"class":302,"line":613},10,[300,615,484],{"class":483},[300,617,487],{"class":313},[300,619,620],{"class":490}," 'select(.type==\"system\" and .subtype==\"compact_boundary\") | .timestamp'",[300,622,494],{"class":313},[300,624,626,628,630,632,634,636,638,640,642,644,646],{"class":302,"line":625},11,[300,627,499],{"class":490},[300,629,503],{"class":502},[300,631,506],{"class":490},[300,633,509],{"class":306},[300,635,512],{"class":502},[300,637,515],{"class":490},[300,639,503],{"class":502},[300,641,520],{"class":490},[300,643,523],{"class":306},[300,645,512],{"class":502},[300,647,648],{"class":490},".jsonl\n",[15,650,651,652,654],{},"If every reply came from one model, the first command prints a single line with that model id and a count. One caveat: Anthropic's sessions docs say the entry format \"is internal to Claude Code and changes between versions.\" Treat these as a quick check, and prefer ",[30,653,32],{}," when it answers the question.",[15,656,657],{},[256,658],{"alt":659,"src":660},"Is Opus 5.5 nerfed diagram: a transcript entry with the message.model field highlighted, next to the /cost usage by model output","/img/blog/is-opus-5-5-nerfed-transcript-model.jpg",[154,662,664],{"id":663},"experiment-flags-in-claudejson-what-they-prove-and-what-they-dont","Experiment flags in ~/.claude.json: what they prove and what they don't",[15,666,667,668,671],{},"A popular theory says Anthropic is A/B testing users into worse behaviour, with the proof in ",[30,669,670],{},"~/.claude.json",". Here's what can be verified.",[15,673,674,675,677,678,680],{},"The docs describe ",[30,676,670],{}," as app state, account details, UI toggles, personal MCP servers and the global config keys ",[30,679,288],{}," writes. They don't document any experiment keys.",[15,682,683],{},"The changelog does confirm Claude Code fetches feature flags and caches them. Version 2.1.214 fixed \"a malformed flag payload\" that \"could wipe the cached feature flags\", and 2.1.300 fixed \"feature flags and staged rollouts staying stale\" behind some proxies. If you open the file you may see cached flag data under internal code names; Anthropic documents none of it, so nothing in it can tell you what a flag does.",[15,685,686],{},"So what does that prove? That your client receives flags, which every staged rollout needs. It doesn't show which flag changes behaviour, or which model the server ran.",[174,688,689],{},[15,690,691,692,694],{},"A cached flag tells you what your client was told. ",[30,693,32],{}," tells you what actually answered. Only one of those is evidence about the model.",[154,696,698],{"id":697},"look-alike-causes-that-feel-like-a-nerf","Look-alike causes that feel like a nerf",[15,700,701],{},"Each of these is documented, and each can make the same model feel worse.",[15,703,704,707,708,710,711,714,715,717,718,720,721,724],{},[25,705,706],{},"A lower effort level than you think."," Opus 5.5 defaults to ",[30,709,102],{},". The docs add that a top-level ",[30,712,713],{},"effortLevel"," in your user settings file \"doesn't count for Opus 5.5\", so a ",[30,716,281],{}," you saved months ago may not apply. Run ",[30,719,96],{},", then ",[30,722,723],{},"/effort high"," if you want it back.",[15,726,727,730,731,734,735,738],{},[25,728,729],{},"Idle compaction."," Since 2.1.299 (14 October) in the terminal, and 2.1.314 (30 October) in VS Code and the desktop app, a conversation that has reached 200,000 tokens and sits idle for about an hour is compacted before its prompt cache expires. You come back to a summary, not the full thread. Look for a line starting ",[30,732,733],{},"Compacted while idle, before the prompt cache expired",", or set ",[30,736,737],{},"\"idleCompaction\": false",". Version 2.1.318 also fixed idle compaction \"replacing a conversation with a reply that was not a summary.\"",[15,740,741,744],{},[25,742,743],{},"Auto-compact at the wrong point."," On 1M-context models, compaction happens late, so the requests before it carry a huge context and the summary afterwards can drop details.",[15,746,747,750],{},[25,748,749],{},"Context bloat."," Every request sends the whole conversation, the docs say. A session holding three finished tasks re-reads all three on every turn of the fourth.",[15,752,753,756,757,760,761,764,765,767],{},[25,754,755],{},"A model switch you made yourself."," ",[30,758,759],{},"You've hit your Opus limit"," and the high-load prompt ",[30,762,763],{},"Opus is experiencing high load, please use /model to switch to Sonnet"," both ask you to switch with ",[30,766,251],{},", and it's easy to forget you did. A switch also starts a fresh prompt cache, so the next request re-reads everything.",[15,769,770],{},[256,771],{"alt":772,"src":773},"Is Opus 5.5 nerfed checklist: six look-alike causes with the setting or command that checks each one","/img/blog/is-opus-5-5-nerfed-look-alikes.jpg",[154,775,777],{"id":776},"so-is-opus-55-nerfed-what-anthropic-has-said","So is Opus 5.5 nerfed? What Anthropic has said",[15,779,780],{},"As of 10 November 2026, Anthropic has published no statement about Opus 5.5 quality in November. We searched Anthropic's news and engineering posts and the help centre, and read the Claude Code changelog through 2.1.322.",[15,782,783],{},"Anthropic's status page shows no Opus 5.5 incident between 28 October and 10 November. The only model incident in that window was \"Degraded performance for Claude Fable 5.1\" on 8 November, from 08:18 to 08:32 UTC, which was resolved.",[15,785,786,787,789,790,792],{},"There is a precedent worth knowing. On 23 April 2026, after a similar wave of complaints, Anthropic published \"An update on recent Claude Code quality reports.\" It traced the drop to three changes in Claude Code itself: a default effort change from ",[30,788,281],{}," to ",[30,791,102],{},", a caching bug that cleared thinking every turn, and a verbosity system prompt. It stated \"The API was not impacted\" and \"We never intentionally degrade our models.\"",[15,794,795],{},"That postmortem doesn't cover November. But the last real regression lived in the tool around the model, in settings you could check. If Anthropic publishes anything on Opus 5.5, we'll add the quote and date here.",[154,797,799],{"id":798},"what-to-do-today","What to do today",[15,801,802],{},"If your Claude Code sessions feel worse, run this in order:",[398,804,805,814,825,835,846],{},[401,806,807,808,810,811,813],{},"Update with ",[30,809,389],{},". Version 2.1.321 fixed a fallback bug, and 2.1.285 fixed ",[30,812,32],{}," model attribution.",[401,815,431,816,818,819,821,822,824],{},[30,817,32],{}," and ",[30,820,83],{},". If a second model appears, run ",[30,823,251],{}," and pick Opus 5.5 again.",[401,826,431,827,829,830,832,833,373],{},[30,828,96],{},". If it says ",[30,831,102],{}," and you want more, run ",[30,834,723],{},[401,836,837,838,840,841,845],{},"Start a fresh session per task with ",[30,839,413],{},", and keep CLAUDE.md lean. Our list of ",[162,842,844],{"href":843},"/blog/skills-that-reduce-token-usage","skills that reduce token usage"," helps with the second part.",[401,847,848,849,851,852,373],{},"If flags keep firing, choose \"Ask each time\" in ",[30,850,288],{}," and test with ",[30,853,268],{},[15,855,856,857,860,861,863],{},"For the one step that needs the strongest model, try Fable: Claude Code 2.1.321 made Fable 5.2 the default Fable model (",[30,858,859],{},"/model fable","). Depending on your plan, Fable can bill to usage credits instead of plan limits, and the ",[30,862,251],{}," picker says \"Requires usage credits\" when it does.",[15,865,866,867,869],{},"Use it for that one step, then run ",[30,868,251],{}," to switch back. At $8 input and $40 output per million tokens on the API, twice Opus 5.5's rates, Fable is priced for hard problems, not routine edits.",[154,871,873],{"id":872},"running-agents-on-a-model-you-choose","Running agents on a model you choose",[15,875,876,877,879],{},"An agent can't notice it was moved to Opus 4.8 at 2am. It can't run ",[30,878,32],{},". It just keeps going on whatever model it ended up with, and you find out from the output.",[15,881,882],{},"That's why we built BetterClaw around your own keys. Your agents run on the provider and model you pick, through BYOK across 28+ providers with zero markup, so the model in your config is the model you pay for.",[15,884,885,886],{},"Give each agent its own key and model: a cheap one for routine checks, a strong one for the hard step. Plans start at $19 a month, and your first workspace starts with a 7-day free trial. ",[162,887,889],{"href":888},"/pricing","See BetterClaw plans",[154,891,893],{"id":892},"check-the-receipts-before-the-verdict","Check the receipts before the verdict",[15,895,896],{},"Feelings about a model are real data, but not evidence on their own.",[15,898,899],{},"Every session leaves receipts: the model that billed each token, the effort it ran at, the moment it compacted, the flag that moved it. Read them first. If they all say Opus 5.5, high effort, full context and it's still worse, you have a report worth filing.",[19,901,903,908],{"type":902},"cta",[15,904,905],{},[25,906,907],{},"Want a second opinion on your agent setup?",[15,909,910,911],{},"If your team runs Claude Code, scripts and agents across several models and you're not sure which one does what, we can map it with you. The AI automation audit is a 30-minute call about how your team works, and within 5 business days you get a written report with where agents fit and how to set them up. ",[162,912,914],{"href":913},"/ai-automation-audit","Book an AI automation audit",[154,916,918],{"id":917},"frequently-asked-questions","Frequently Asked Questions",[920,921,27],"h3",{"id":922},"is-opus-55-nerfed",[15,924,925,926,928,929,931,932,818,934,936],{},"There is no public evidence that Anthropic changed Opus 5.5, and Anthropic has made no statement about it as of 10 November 2026. Many reported drops trace to documented Claude Code behaviour instead: a safeguards fallback to Opus 4.8 or Opus 5, the ",[30,927,102],{}," default effort, idle compaction or context bloat. Check ",[30,930,32],{},", ",[30,933,83],{},[30,935,96],{}," first.",[920,938,940],{"id":939},"how-do-i-check-which-claude-model-is-running","How do I check which Claude model is running?",[15,942,431,943,456,945,947,948,950,951,953,954,957],{},[30,944,83],{},[30,946,32],{}," (an alias for ",[30,949,69],{},") to see every model that billed tokens in the session. For a reply-by-reply view, the ",[30,952,112],{}," field in your session transcript under ",[30,955,956],{},"~/.claude/projects/"," names the model, though Anthropic says that file format is internal and can change.",[920,959,961],{"id":960},"what-does-safeguards-stopped-the-response-mean","What does safeguards stopped the response mean?",[15,963,964,965,968,969,972,973,975],{},"The full line reads ",[30,966,967],{},"Opus 5.5's safeguards stopped the response above · continuing once with that noted",". It means the named model's safety classifiers stopped a response, and the error that follows gives the category, such as ",[30,970,971],{},"[cyber]",". Press Esc twice or run ",[30,974,406],{}," to rephrase, and update Claude Code: 2.1.321 fixed a fallback bug that mentioned stopped responses.",[920,977,979],{"id":978},"why-does-claude-code-switch-to-opus-48","Why does Claude Code switch to Opus 4.8?",[15,981,982,983,985,986,988,989,789,992,373],{},"When Opus 5.5's cybersecurity classifiers flag a request, Claude Code re-runs it on Opus 4.8, and biology flags re-run on Opus 5. The session then stays on the fallback model until you run ",[30,984,251],{},". To be asked first, choose Ask each time in ",[30,987,288],{}," or set ",[30,990,991],{},"switchModelsOnFlag",[30,993,994],{},"false",[920,996,998],{"id":997},"is-fable-51-better-than-opus-55-right-now","Is Fable 5.1 better than Opus 5.5 right now?",[15,1000,1001],{},"Fable 5.1 is no longer the default Fable model: Claude Code 2.1.321 made Fable 5.2 the default Fable model on 9 November. Anthropic positions Fable for the hardest, longest-running tasks, and its API price is higher ($8 input and $40 output per million tokens for Fable 5.2, versus $4 and $20 for Opus 5.5). Depending on your plan, Fable usage can bill to usage credits.",[920,1003,1005],{"id":1004},"how-do-i-stop-claude-code-from-switching-models-when-a-message-is-flagged","How do I stop Claude Code from switching models when a message is flagged?",[15,1007,431,1008,1010,1011,789,1013,1015,1016,1018],{},[30,1009,288],{},", select Switch models when a message is flagged, and choose Ask each time, or set ",[30,1012,991],{},[30,1014,994],{}," in your settings file. To turn the automatic switch off completely, set the environment variable ",[30,1017,332],{},". A flagged request then ends with the refusal instead of moving your session.",[1020,1021,1022],"style",{},"html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .sAwPA, html code.shiki .sAwPA{--shiki-default:#6A737D}html pre.shiki code .s7eDp, html code.shiki .s7eDp{--shiki-default:#6F42C1}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html pre.shiki code .sD7c4, html code.shiki .sD7c4{--shiki-default:#D73A49}",{"title":296,"searchDepth":310,"depth":310,"links":1024},[1025,1026,1027,1028,1029,1030,1031,1032,1033,1034,1035],{"id":156,"depth":310,"text":157},{"id":181,"depth":310,"text":182},{"id":341,"depth":310,"text":342},{"id":427,"depth":310,"text":428},{"id":663,"depth":310,"text":664},{"id":697,"depth":310,"text":698},{"id":776,"depth":310,"text":777},{"id":798,"depth":310,"text":799},{"id":872,"depth":310,"text":873},{"id":892,"depth":310,"text":893},{"id":917,"depth":310,"text":918,"children":1036},[1037,1038,1039,1040,1041,1042],{"id":922,"depth":323,"text":27},{"id":939,"depth":323,"text":940},{"id":960,"depth":323,"text":961},{"id":978,"depth":323,"text":979},{"id":997,"depth":323,"text":998},{"id":1004,"depth":323,"text":1005},"Troubleshooting","2026-11-10","Is Opus 5.5 nerfed? How to check which model your Claude Code session used, what the safeguards fallback does and what Anthropic has said.","md",false,"/img/blog/is-opus-5-5-nerfed.jpg","Is Opus 5.5 nerfed cover: a model card with a question mark and a magnifying glass over the Opus 5.5 safeguards log line",512,1024,null,{},"/blog/is-opus-5-5-nerfed","9 min read",{"title":5,"description":1045},"Is Opus 5.5 Nerfed? How to Check Which Model Actually Ran","blog/is-opus-5-5-nerfed",[1060,1061,1062,1063,1064,991,1065],"is opus 5.5 nerfed","opus 5.5 nerfed","claude code which model is it using","opus 5.5 safeguards stopped the response","claude code switch to opus 4.8","claude code check model","7I54KYpTyYdpUh9vqif3jNTdufmraH-8azQdwnuvtz0",[1068,1075,1083],{"stem":1069,"title":1070,"description":1071,"date":1044,"image":1072,"category":1043,"redirected":1047,"author":1073,"readingTime":1074},"blog/openclaw-fallback-model-not-working","OpenClaw Fallback Model Not Working After 2026.10.10: Why","OpenClaw fallback model not working after 2026.10.10? Scheduled jobs now stay on one provider by design. What still works, the config and 3 workarounds.","/img/blog/openclaw-fallback-model-not-working.jpg",{"name":7,"role":8,"avatar":9,"bio":10},"8 min read",{"stem":1076,"title":1077,"description":1078,"date":1079,"image":1080,"category":1081,"redirected":1047,"author":1082,"readingTime":1055},"blog/gpt-6-1-terra-vs-gpt-6-astra-agents","GPT-6.1 Terra vs GPT-6 Astra: Is the Cheaper Model Enough for Your Agent?","GPT-6.1 Terra vs GPT-6 Astra for AI agents: API pricing, computer use, visual reasoning and a real cost example. When the cheaper model is enough.","2026-11-09","/img/blog/gpt-6-1-terra-vs-gpt-6-astra-agents.jpg","Comparison",{"name":7,"role":8,"avatar":9,"bio":10},{"stem":1084,"title":1085,"description":1086,"date":1079,"image":1087,"category":1088,"redirected":1047,"author":1089,"readingTime":1090},"blog/mistral-large-4-hardware-requirements","Mistral Large 4 Hardware Requirements: Can You Run a 1T Model Locally?","Mistral Large 4 hardware requirements: memory at each quantisation, which Macs, GPUs and workstations run it, build costs and when the API is cheaper.","/img/blog/mistral-large-4-hardware-requirements.jpg","Guides",{"name":7,"role":8,"avatar":9,"bio":10},"10 min read",1794321828905]