A rambling comment: I think this is the first time we've had a third minor version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massive capability leaps, eg, Sonnet 3.5, Opus 4.5). So now the Opus 4.5 family has successors 4.6, 4.7, and 4.8, each posting fairly modest claimed gains. My own experience w/ 4.6 and 4.7 are that I don't firmly grasp any capabilities improvements over my memory of 4.5, but it's all so fuzzy that it's truly difficult to tell. Maybe my own tastes are saturated now (it's smarter than me?) and I'll never again perceive model progress. Maybe the incrementalism is such that I'd notice immediately if my 4.7 workflows were redirected now to 4.5. Difficult spot for the labs to be in because, if they have a stronger product, I'd prefer they release it and that I can use it. But as this dynamic continues, the improvements are going to be less and less legible for end-users, who will complain about the churn-without-payoff, even when the payoff may actually be real.
痛点分析发布于 2026/05/28
痛点为 AI 基于上游原始证据的初步提炼；未包含额外中国市场检索。
痛点
在Hacker News关于Claude Opus 4.8的讨论中，用户NiloCK指出，模型版本频繁小幅度更新（4.5→4.6→4.7→4.8）但改进难以感知，导致用户无法清晰判断能力提升，甚至怀疑自己的感知是否饱和。这种“增量无感”的迭代模式让用户面临选择困难：他们需要决定是否升级或切换工作流，但缺乏可量化的证据来支撑决策。同时，用户colonCapitalDee提到“自适应思考”功能无法可靠触发，导致模型输出质量不稳定，用户不得不手动干预或接受次优结果。这些痛点共同造成了时间浪费（反复测试版本差异）、决策延迟（无法确定最佳模型）以及心理负担（对模型能力的不信任感）。
External Article
External article summary
Our latest model, Claude Opus 4.8, is an upgrade to our Opus class of models, with stronger performance across coding, agentic tasks, and professional work, and the consistency to handle long-running work.
External Article
External article source
- Article title
- Introducing Claude Opus 4.8
- Host
- www.anthropic.com
§ Dossier
Selected HN comments
NiloCK
"Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor." This is a refreshing attitude! I've also verified that you can now turn off adaptive thinking in the web UI, which is great. I've had a lot of problems with thinking not triggering and the model producing sub-par output. Glad we can finally turn it off. (I hope being able to turn off adaptive thinking is new, if I could have turned it off at any time that would be embarrassing)
colonCapitalDee
> One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest On the contrary, they appear trained to say "Honestly" or "I have to be transparent with you" at inverse proportion to certainty. Put another way, if they are certain, they don't use "Honestly", and if they are just wrong, or know they don't know, they don't use "Honestly". They use "honestly" on the bubble, to the degree it's a tell that whatever it's asserting or doing is shakily grounded, sketchy or lazy work, or a host of other reasons you shouldn't trust it. This training seems instead to be making it performatively punch up claims it cannot substantiate. See also: https://news.ycombinator.com/item?id=48312182
Terretta
> Not only that, but we plan to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before they can be generally released. We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks. Probably more interesting than the 4.8 release.
northern-lights
I generated pelicans riding bicycles on both thinking level low and thinking level high: https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304... The high one is notably better - the bicycle frame is the correct shape, unlike thinking level low. For comparison, here's Opus 4.7: https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c...
simonw
源数据· Raw Archive
- source
- Hacker News
- upstream_source
- hacker_news
- upstream_item_id
- 48311647
- daily_ranking_item_id
- c46fd610-b113-4b80-96d8-c80469b89462
- rank_date
- 2026-05-29
- rank
- 1
- name
- Claude Opus 4.8
- tagline
- www.anthropic.com
- votes_count
- 1,026
- comments_count
- 812
- created_at_on_source
- 2026-05-28T16:49:14.000Z
media / source-specific data
{
"author": "craigmart",
"hn_item_id": 48311647,
"external_url": "https://www.anthropic.com/news/claude-opus-4-8"
}raw_payload
{
"by": "craigmart",
"id": 48311647,
"url": "https://www.anthropic.com/news/claude-opus-4-8",
"kids": [
48311998,
48311843,
48316016,
48311816,
48311979,
48313432,
48314967,
48311777,
48311934,
48312609,
48312774,
48313543,
48312135,
48315239,
48311780,
48312684,
48315952,
48311832,
48312633,
48312067,
48312255,
48312380,
48313544,
48312224,
48312922,
48313098,
48312333,
48314830,
48314414,
48312984,
48311823,
48313986,
48313915,
48312686,
48312132,
48314329,
48311957,
48311870,
48315773,
48313299,
48314360,
48311851,
48315530,
48312155,
48312381,
48312738,
48313121,
48311944,
48312157,
48311730,
48315223,
48314433,
48311740,
48315124,
48312361,
48312422,
48311726,
48314882,
48311814,
48311967,
48312231,
48312791,
48311971,
48314661,
48313578,
48312904,
48311958,
48312571,
48311984,
48312291,
48312119,
48312028,
48314405,
48312500,
48312386,
48312274,
48313648,
48311873,
48312225,
48313168,
48315861,
48314255,
48311846,
48313146,
48315867,
48311801,
48311945,
48312859,
48312472,
48314057,
48312075,
48311708,
48313336,
48312178,
48311798,
48312366,
48313137,
48311811,
48313899,
48313337,
48312329,
48312317,
48313293,
48312215,
48314780,
48313280,
48314814,
48314788,
48311937,
48311918,
48314186,
48312808,
48311890,
48313413,
48312554,
48311732,
48315664,
48311881,
48311833,
48311802,
48311747,
48313361,
48313141,
48312816,
48315727,
48312428,
48311830,
48311731,
48311897,
48313769,
48312552,
48312057,
48311702,
48312280,
48315364,
48314466,
48315192,
48312811,
48312130,
48314925,
48312661,
48311844,
48311717,
48311790,
48311737,
48311895
],
"time": 1779986954,
"type": "story",
"score": 1026,
"title": "Claude Opus 4.8",
"descendants": 812
}source_raw_snapshot
{
"id": "02f8f186-dc52-4b16-bdd5-f5b11dbf8e82",
"daily_ranking_item_id": "c46fd610-b113-4b80-96d8-c80469b89462",
"source": "hacker_news",
"external_id": "48311647",
"fetched_at": "2026-05-28T22:01:23.715Z",
"story_raw": {
"by": "craigmart",
"id": 48311647,
"url": "https://www.anthropic.com/news/claude-opus-4-8",
"kids": [
48311998,
48311843,
48316016,
48311816,
48311979,
48313432,
48314967,
48311777,
48311934,
48312609,
48312774,
48313543,
48312135,
48315239,
48311780,
48312684,
48315952,
48311832,
48312633,
48312067,
48312255,
48312380,
48313544,
48312224,
48312922,
48313098,
48312333,
48314830,
48314414,
48312984,
48311823,
48313986,
48313915,
48312686,
48312132,
48314329,
48311957,
48311870,
48315773,
48313299,
48314360,
48311851,
48315530,
48312155,
48312381,
48312738,
48313121,
48311944,
48312157,
48311730,
48315223,
48314433,
48311740,
48315124,
48312361,
48312422,
48311726,
48314882,
48311814,
48311967,
48312231,
48312791,
48311971,
48314661,
48313578,
48312904,
48311958,
48312571,
48311984,
48312291,
48312119,
48312028,
48314405,
48312500,
48312386,
48312274,
48313648,
48311873,
48312225,
48313168,
48315861,
48314255,
48311846,
48313146,
48315867,
48311801,
48311945,
48312859,
48312472,
48314057,
48312075,
48311708,
48313336,
48312178,
48311798,
48312366,
48313137,
48311811,
48313899,
48313337,
48312329,
48312317,
48313293,
48312215,
48314780,
48313280,
48314814,
48314788,
48311937,
48311918,
48314186,
48312808,
48311890,
48313413,
48312554,
48311732,
48315664,
48311881,
48311833,
48311802,
48311747,
48313361,
48313141,
48312816,
48315727,
48312428,
48311830,
48311731,
48311897,
48313769,
48312552,
48312057,
48311702,
48312280,
48315364,
48314466,
48315192,
48312811,
48312130,
48314925,
48312661,
48311844,
48311717,
48311790,
48311737,
48311895
],
"time": 1779986954,
"type": "story",
"score": 1026,
"title": "Claude Opus 4.8",
"descendants": 812
},
"stats_raw": {
"time": 1779986954,
"score": 1026,
"descendants": 812
},
"aux_raw": {
"external_url": "https://www.anthropic.com/news/claude-opus-4-8",
"hn_comment_url": "https://news.ycombinator.com/item?id=48311647",
"normalized_text": null,
"external_article": {
"title": "Introducing Claude Opus 4.8",
"excerpt": "We’re upgrading Claude Opus to a new version: Claude Opus 4.8. It builds on Opus 4.7 with improvements across benchmarks, and is a more effective collaborator. It’s available today for the same price.\n\nOpus 4.8 launches alongside several new features. Users on claude.ai now have control over the amount of effort Claude puts into a task. Claude Code has a new “dynamic workflows” feature that allows it to tackle very large-scale problems. And fast mode for Opus 4.8—where the model can work at 2.5× the speed—is now three times cheaper than it was for previous models.\n\nThe table below shows how Opus 4.8 compares to its predecessor and to other models on tests of coding, agentic skills, reasoning, and practical knowledge work tasks. More details and a much wider range of capability evaluations are provided in the Claude Opus 4.8 System Card .\n\nEarly testers have found Claude Opus 4.8 to be more reliable and sharper in its judgement when it’s performing agentic tasks. Below are quotes from many of these testers about their experience collaborating with Opus 4.8:\n\nOne of the most prominent improvements in Opus 4.8 is its honesty . We train all our models to be honest—for instance, to avoi",
"final_url": "https://www.anthropic.com/news/claude-opus-4-8",
"fetched_at": "2026-05-28T22:01:21.556Z",
"description": "Our latest model, Claude Opus 4.8, is an upgrade to our Opus class of models, with stronger performance across coding, agentic tasks, and professional work, and the consistency to handle long-running work."
},
"selected_comments": [
{
"id": 48311998,
"raw": {
"by": "NiloCK",
"id": 48311998,
"kids": [
48312244,
48312185,
48312242,
48314086,
48312934,
48312195,
48313284,
48312229,
48312567,
48315207,
48313581,
48315438,
48312205,
48313886,
48314669,
48314032,
48312548,
48315280,
48312216,
48312276,
48314390,
48313797,
48315644,
48313793,
48315279,
48315186,
48314147,
48312343,
48312117,
48314116,
48314140,
48313688
],
"text": "A rambling comment:<p>I think this is the first time we've had a third <i>minor</i> version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massive capability leaps, eg, Sonnet 3.5, Opus 4.5).<p>So now the Opus 4.5 family has successors 4.6, 4.7, and 4.8, each posting fairly modest claimed gains. My own experience w/ 4.6 and 4.7 are that I don't <i>firmly grasp</i> any capabilities improvements over my memory of 4.5, but it's all so fuzzy that it's truly difficult to tell.<p>Maybe my own tastes are saturated now (it's smarter than me?) and I'll never again perceive model progress. Maybe the incrementalism is such that I'd notice immediately if my 4.7 workflows were redirected now to 4.5.<p>Difficult spot for the labs to be in because, if they have a stronger product, I'd prefer they release it and that I can use it.<p>But as this dynamic continues, the improvements are going to be less and less legible for end-users, who will complain about the churn-without-payoff, even when the payoff may actually be real.",
"time": 1779988006,
"type": "comment",
"parent": 48311647
},
"body": "A rambling comment: I think this is the first time we've had a third minor version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massive capability leaps, eg, Sonnet 3.5, Opus 4.5). So now the Opus 4.5 family has successors 4.6, 4.7, and 4.8, each posting fairly modest claimed gains. My own experience w/ 4.6 and 4.7 are that I don't firmly grasp any capabilities improvements over my memory of 4.5, but it's all so fuzzy that it's truly difficult to tell. Maybe my own tastes are saturated now (it's smarter than me?) and I'll never again perceive model progress. Maybe the incrementalism is such that I'd notice immediately if my 4.7 workflows were redirected now to 4.5. Difficult spot for the labs to be in because, if they have a stronger product, I'd prefer they release it and that I can use it. But as this dynamic continues, the improvements are going to be less and less legible for end-users, who will complain about the churn-without-payoff, even when the payoff may actually be real.",
"is_op": false,
"author": "NiloCK",
"raw_body": "A rambling comment:<p>I think this is the first time we've had a third <i>minor</i> version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massive capability leaps, eg, Sonnet 3.5, Opus 4.5).<p>So now the Opus 4.5 family has successors 4.6, 4.7, and 4.8, each posting fairly modest claimed gains. My own experience w/ 4.6 and 4.7 are that I don't <i>firmly grasp</i> any capabilities improvements over my memory of 4.5, but it's all so fuzzy that it's truly difficult to tell.<p>Maybe my own tastes are saturated now (it's smarter than me?) and I'll never again perceive model progress. Maybe the incrementalism is such that I'd notice immediately if my 4.7 workflows were redirected now to 4.5.<p>Difficult spot for the labs to be in because, if they have a stronger product, I'd prefer they release it and that I can use it.<p>But as this dynamic continues, the improvements are going to be less and less legible for end-users, who will complain about the churn-without-payoff, even when the payoff may actually be real.",
"created_at": 1779988006,
"reply_count": 32
},
{
"id": 48311843,
"raw": {
"by": "colonCapitalDee",
"id": 48311843,
"kids": [
48316024,
48314419,
48315552,
48315433,
48312319,
48312349,
48313301,
48312939,
48314699,
48312948,
48312976,
48313541
],
"text": ""Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor."<p>This is a refreshing attitude!<p>I've also verified that you can now turn off adaptive thinking in the web UI, which is great. I've had a lot of problems with thinking not triggering and the model producing sub-par output. Glad we can finally turn it off. (I hope being able to turn off adaptive thinking is new, if I could have turned it off at any time that would be embarrassing)",
"time": 1779987503,
"type": "comment",
"parent": 48311647
},
"body": "\"Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor.\" This is a refreshing attitude! I've also verified that you can now turn off adaptive thinking in the web UI, which is great. I've had a lot of problems with thinking not triggering and the model producing sub-par output. Glad we can finally turn it off. (I hope being able to turn off adaptive thinking is new, if I could have turned it off at any time that would be embarrassing)",
"is_op": false,
"author": "colonCapitalDee",
"raw_body": ""Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor."<p>This is a refreshing attitude!<p>I've also verified that you can now turn off adaptive thinking in the web UI, which is great. I've had a lot of problems with thinking not triggering and the model producing sub-par output. Glad we can finally turn it off. (I hope being able to turn off adaptive thinking is new, if I could have turned it off at any time that would be embarrassing)",
"created_at": 1779987503,
"reply_count": 12
},
{
"id": 48316016,
"raw": {
"by": "Terretta",
"id": 48316016,
"text": "> <i>One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest</i><p>On the contrary, they appear trained to say "Honestly" or "I have to be transparent with you" at inverse proportion to certainty.<p>Put another way, if they are certain, they don't use "Honestly", and if they are just wrong, or know they don't know, they don't use "Honestly".<p>They use "honestly" on the bubble, to the degree it's a tell that whatever it's asserting or doing is shakily grounded, sketchy or lazy work, or a host of other reasons you shouldn't trust it.<p>This training seems instead to be making it performatively punch up claims it cannot substantiate.<p>See also: <a href=\"https://news.ycombinator.com/item?id=48312182\">https://news.ycombinator.com/item?id=48312182</a>",
"time": 1780005048,
"type": "comment",
"parent": 48311647
},
"body": "> One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest On the contrary, they appear trained to say \"Honestly\" or \"I have to be transparent with you\" at inverse proportion to certainty. Put another way, if they are certain, they don't use \"Honestly\", and if they are just wrong, or know they don't know, they don't use \"Honestly\". They use \"honestly\" on the bubble, to the degree it's a tell that whatever it's asserting or doing is shakily grounded, sketchy or lazy work, or a host of other reasons you shouldn't trust it. This training seems instead to be making it performatively punch up claims it cannot substantiate. See also: https://news.ycombinator.com/item?id=48312182",
"is_op": false,
"author": "Terretta",
"raw_body": "> <i>One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest</i><p>On the contrary, they appear trained to say "Honestly" or "I have to be transparent with you" at inverse proportion to certainty.<p>Put another way, if they are certain, they don't use "Honestly", and if they are just wrong, or know they don't know, they don't use "Honestly".<p>They use "honestly" on the bubble, to the degree it's a tell that whatever it's asserting or doing is shakily grounded, sketchy or lazy work, or a host of other reasons you shouldn't trust it.<p>This training seems instead to be making it performatively punch up claims it cannot substantiate.<p>See also: <a href=\"https://news.ycombinator.com/item?id=48312182\">https://news.ycombinator.com/item?id=48312182</a>",
"created_at": 1780005048,
"reply_count": 0
},
{
"id": 48311816,
"raw": {
"by": "northern-lights",
"id": 48311816,
"kids": [
48314935,
48313036,
48314805,
48312604,
48312103,
48312743
],
"text": "> Not only that, but we plan to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before they can be generally released. We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks.<p>Probably more interesting than the 4.8 release.",
"time": 1779987427,
"type": "comment",
"parent": 48311647
},
"body": "> Not only that, but we plan to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before they can be generally released. We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks. Probably more interesting than the 4.8 release.",
"is_op": false,
"author": "northern-lights",
"raw_body": "> Not only that, but we plan to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before they can be generally released. We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks.<p>Probably more interesting than the 4.8 release.",
"created_at": 1779987427,
"reply_count": 6
},
{
"id": 48311979,
"raw": {
"by": "simonw",
"id": 48311979,
"kids": [
48313081,
48315019,
48314391,
48312275,
48315889,
48312466,
48312482,
48312320,
48313089,
48312173,
48313700,
48315228,
48312909,
48312237,
48312105,
48313545,
48313009,
48312079
],
"text": "I generated pelicans riding bicycles on both thinking level low and thinking level high:<p><a href=\"https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304dc6?permalink_comment_id=6172953#gistcomment-6172953\" rel=\"nofollow\">https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304...</a><p>The high one is notably better - the bicycle frame is the correct shape, unlike thinking level low.<p>For comparison, here's Opus 4.7: <a href=\"https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c118?permalink_comment_id=6104087#gistcomment-6104087\" rel=\"nofollow\">https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c...</a>",
"time": 1779987960,
"type": "comment",
"parent": 48311647
},
"body": "I generated pelicans riding bicycles on both thinking level low and thinking level high: https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304... The high one is notably better - the bicycle frame is the correct shape, unlike thinking level low. For comparison, here's Opus 4.7: https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c...",
"is_op": false,
"author": "simonw",
"raw_body": "I generated pelicans riding bicycles on both thinking level low and thinking level high:<p><a href=\"https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304dc6?permalink_comment_id=6172953#gistcomment-6172953\" rel=\"nofollow\">https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304...</a><p>The high one is notably better - the bicycle frame is the correct shape, unlike thinking level low.<p>For comparison, here's Opus 4.7: <a href=\"https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c118?permalink_comment_id=6104087#gistcomment-6104087\" rel=\"nofollow\">https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c...</a>",
"created_at": 1779987960,
"reply_count": 18
}
],
"presentation_fields": {
"title": "Claude Opus 4.8",
"tagline": "www.anthropic.com",
"website_url": "https://www.anthropic.com/news/claude-opus-4-8",
"canonical_url": "https://news.ycombinator.com/item?id=48311647"
},
"external_url_hostname": "www.anthropic.com",
"selected_comments_raw": [
{
"by": "NiloCK",
"id": 48311998,
"kids": [
48312244,
48312185,
48312242,
48314086,
48312934,
48312195,
48313284,
48312229,
48312567,
48315207,
48313581,
48315438,
48312205,
48313886,
48314669,
48314032,
48312548,
48315280,
48312216,
48312276,
48314390,
48313797,
48315644,
48313793,
48315279,
48315186,
48314147,
48312343,
48312117,
48314116,
48314140,
48313688
],
"text": "A rambling comment:<p>I think this is the first time we've had a third <i>minor</i> version bump on a frontier Anthropic model. (I count the 0.5s as major here, because they've been issued non-sequentially and also corresponded to massive capability leaps, eg, Sonnet 3.5, Opus 4.5).<p>So now the Opus 4.5 family has successors 4.6, 4.7, and 4.8, each posting fairly modest claimed gains. My own experience w/ 4.6 and 4.7 are that I don't <i>firmly grasp</i> any capabilities improvements over my memory of 4.5, but it's all so fuzzy that it's truly difficult to tell.<p>Maybe my own tastes are saturated now (it's smarter than me?) and I'll never again perceive model progress. Maybe the incrementalism is such that I'd notice immediately if my 4.7 workflows were redirected now to 4.5.<p>Difficult spot for the labs to be in because, if they have a stronger product, I'd prefer they release it and that I can use it.<p>But as this dynamic continues, the improvements are going to be less and less legible for end-users, who will complain about the churn-without-payoff, even when the payoff may actually be real.",
"time": 1779988006,
"type": "comment",
"parent": 48311647
},
{
"by": "colonCapitalDee",
"id": 48311843,
"kids": [
48316024,
48314419,
48315552,
48315433,
48312319,
48312349,
48313301,
48312939,
48314699,
48312948,
48312976,
48313541
],
"text": ""Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor."<p>This is a refreshing attitude!<p>I've also verified that you can now turn off adaptive thinking in the web UI, which is great. I've had a lot of problems with thinking not triggering and the model producing sub-par output. Glad we can finally turn it off. (I hope being able to turn off adaptive thinking is new, if I could have turned it off at any time that would be embarrassing)",
"time": 1779987503,
"type": "comment",
"parent": 48311647
},
{
"by": "Terretta",
"id": 48316016,
"text": "> <i>One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest</i><p>On the contrary, they appear trained to say "Honestly" or "I have to be transparent with you" at inverse proportion to certainty.<p>Put another way, if they are certain, they don't use "Honestly", and if they are just wrong, or know they don't know, they don't use "Honestly".<p>They use "honestly" on the bubble, to the degree it's a tell that whatever it's asserting or doing is shakily grounded, sketchy or lazy work, or a host of other reasons you shouldn't trust it.<p>This training seems instead to be making it performatively punch up claims it cannot substantiate.<p>See also: <a href=\"https://news.ycombinator.com/item?id=48312182\">https://news.ycombinator.com/item?id=48312182</a>",
"time": 1780005048,
"type": "comment",
"parent": 48311647
},
{
"by": "northern-lights",
"id": 48311816,
"kids": [
48314935,
48313036,
48314805,
48312604,
48312103,
48312743
],
"text": "> Not only that, but we plan to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. Models of this capability level require stronger cyber safeguards before they can be generally released. We’re making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks.<p>Probably more interesting than the 4.8 release.",
"time": 1779987427,
"type": "comment",
"parent": 48311647
},
{
"by": "simonw",
"id": 48311979,
"kids": [
48313081,
48315019,
48314391,
48312275,
48315889,
48312466,
48312482,
48312320,
48313089,
48312173,
48313700,
48315228,
48312909,
48312237,
48312105,
48313545,
48313009,
48312079
],
"text": "I generated pelicans riding bicycles on both thinking level low and thinking level high:<p><a href=\"https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304dc6?permalink_comment_id=6172953#gistcomment-6172953\" rel=\"nofollow\">https://gist.github.com/simonw/68560eddb0b268a8417f80ceb7304...</a><p>The high one is notably better - the bicycle frame is the correct shape, unlike thinking level low.<p>For comparison, here's Opus 4.7: <a href=\"https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c118?permalink_comment_id=6104087#gistcomment-6104087\" rel=\"nofollow\">https://gist.github.com/simonw/afcb19addf3f38eb1996e1ebe749c...</a>",
"time": 1779987960,
"type": "comment",
"parent": 48311647
}
]
},
"selection_meta": {
"discussion_depth": "top_comments_v1",
"external_article": {
"status": "ok",
"final_url": "https://www.anthropic.com/news/claude-opus-4-8",
"status_code": 200,
"content_type": "text/html; charset=utf-8",
"failure_reason": null
},
"snapshot_version": "hn_story_v3",
"selected_comments_count": 5,
"external_article_resolved": true,
"text_normalization_applied": false
},
"created_at": "2026-05-28T22:01:23.765Z",
"updated_at": "2026-05-28T22:01:23.765Z"
}