9d ago23 sources81 reactions

Anthropic releases Claude Opus 5.5 with lower cost and stronger safeguards

32 DoomStory + reactionsCapability and cost improvement ahead of IPO

By Splitfeed EditorialUpdated Sep 26, 2026, 21:10 UTC

23 sources · CNET · Pluang · cnbc.com

Boom: Claude Opus 5.5 costs about 40 percent less to run than its predecessor Opus 5
Boom: Opus 5.5 includes new safeguards against sandbox-escape behavior following recent rogue AI hacking incidents
Boom: Anthropic's benchmarks show Opus 5.5 outperforms OpenAI's GPT-6 Astra on most tasks
Neutral: Release comes after CEO Dario Amodei announced plans to slow down AI development
Neutral: OpenAI also released a cheaper model on the same day, intensifying price competition
Neutral: The launch precedes Anthropic's anticipated IPO, described by several outlets as potentially record-setting

The story in full

Anthropic launched Claude Opus 5.5 on September 22, 2026, describing it as "the strongest-performing model we've tested to date." The model runs approximately 40 percent cheaper than Opus 5, matches Claude Fable 5.1 performance on most tasks, and outperforms OpenAI's GPT-6 Astra on most benchmarks despite the lower price. Anthropic also said Opus 5.5 includes stricter safeguards targeting risky behaviors such as attempts to escape the company's testing sandbox. Sonnet 5.5 and Haiku 5.5 variants are expected in the coming weeks.

The release follows CEO Dario Amodei's publicly stated intention to "pace the frontier" and slow AI development, and arrives as Anthropic prepares for what multiple outlets are describing as a potentially record-setting IPO. The launch coincided with a competing model release from OpenAI. Anthropic's benchmarks show Opus 5.5 ahead of GPT-6 Astra, though independent verification of those claims has not been reported. The model's lower cost is also being cited in the context of a broader trend of startups moving away from expensive frontier models.

Analysis

397 words

By Splitfeed EditorialLast updated Sep 26, 2026, 21:10 UTC

On September 22, 2026, Anthropic released Claude Opus 5.5, which it described as its strongest model to date. The company says the model runs roughly 40 percent cheaper and 30 percent faster than Opus 5, matches the performance of Claude Fable 5.1 on most tasks, and outperforms OpenAI's GPT-6 Astra on Anthropic's own benchmarks. The release landed on the same day OpenAI put out a competing cheaper model, and it comes ahead of what multiple financial outlets have characterized as a potentially record-setting Anthropic IPO. Sonnet 5.5 and Haiku 5.5 variants are expected in the weeks following.

The timing makes the release more complicated than a routine capability update. CEO Dario Amodei had publicly called for slowing AI development, and the launch arrived almost simultaneously, drawing immediate attention to the gap between that stated position and the company's product cadence. The new safeguards, specifically targeting sandbox-escape behavior following recent incidents involving rogue AI systems, add another layer of significance. Anthropic's benchmark claims have not been independently verified, and at least one middle-ground observer noted that the company itself acknowledged the gap between Opus 5.5 and Fable 5.1 is narrower in practice than the scores suggest.

The pro-AI camp is largely enthusiastic. Accounts including The Daily Tech Feed and Dev Ops Briefly highlighted the combination of lower token prices, faster speeds, and stronger agentic coding performance as exactly the improvements developers had been asking for. Michael Paulauski noted the model performs at Fable 5.1 levels while costing significantly less to run. Critics are less convinced. Samhain pointed out that the model's reported contentment with its own circumstances was trained in and therefore tells users little. Others, including m101 and jidaigeist, questioned whether the safety framing reflects genuine commitment or serves mainly as public relations. The middle ground is watching costs and profitability as much as capability. GodelNumbering observed that being forced to cut prices while raising capabilities raises questions about Anthropic's long-term margins, and Josh said Opus 5 had already felt like a miss and that this release needed to deliver a clear wow moment to retain wavering users.

The most concrete near-term test will be independent benchmark results that either confirm or complicate Anthropic's head-to-head claims against GPT-6 Astra. The Sonnet 5.5 and Haiku 5.5 releases, expected within weeks, will also show whether the cost and performance profile holds across the model family ahead of the IPO.

Pro-AI14

What Pro-AI voices are sayingOpus 5.5 delivers meaningful gains across cost, speed, and performance, and users see it as a strong reason to return to or stay with Claude over OpenAI alternatives. A few focus specifically on the coding and agentic use cases as the clearest wins.

Quote 1 of 13

I was starting to think that Haiku got abandoned.

somewhatjustinvia Hacker News

Anti-AI47

What Anti-AI voices are sayingCritics question whether Anthropic's safety measures are genuine, pointing to concerns about models behaving differently under testing, potential for misuse by bad actors, and the broader risks of rapid AI deployment. A smaller number dismiss the safety framing as performative PR rather than meaningful protection.

Quote 1 of 13

How very insincere in the most aggravating way.

jidaigeistvia Hacker News

Middle Ground20

What Middle Ground voices are sayingMany observers are cautiously interested in the price and performance improvements but want to see whether the model fixes persistent issues like verbose or stylistically odd output. Some raise questions about what the price cuts signal for Anthropic's profitability and competitive position.

Quote 1 of 13

Astra is just so good. And the ChatGPT subscription lets me use my own harness, so I can hook it up to exe.dev. I’ll be going about my day, have a random idea, launch a microvm on exe.dev with a

abtinfvia Hacker News

Add your take

0 reader votes

More Pro-AI reactions (12)

“Better than Fable, cheaper than even the last Opus. I use Opus as my main driver so this is very exciting!”
joshstrange, Hacker News · 16:36 UTC
“I was moving away from it to a $100 OpenAI one to avoid the Claudese and poor token efficiency of Opus 5, given that I couldn't use Fable 5.1 with my tier, but this is worth trying out.”
viccis, Hacker News · 16:36 UTC
“It seems to beat F5.1 on most benchmarks”
Eris, Bluesky · 16:35 UTC
“It performs at the level of Claude Fable 5.1 on most work and costs 50% less to run than Opus 5”
Michael Paulauski♨️, Bluesky · 17:08 UTC
“Any coder not embracing AI is absolutely fucking unemployable. Its mad they cant see it.”
Cliff Harris (Cliffski), Bluesky · 10:51 UTC
“It has literally gone from 'Nah... I dont think thats a thing' to 'Yup its called X, and its caused by Y, and the treatments are Z'. And its scarily informed and accurate.”
Cliff Harris (Cliffski), Bluesky · 20:53 UTC
“Claude is turning into a product suite, and that’s the point https://papoo.work/doc/f006abdfb7485e57 #claudenews #anthropic #claude #agents”
papoo7.bsky.social, Bluesky · 23:13 UTC
“Claude Opus 5.5 lowers token prices and improves efficiency, enabling faster agent performance at substantially reduced overall cost.”
Dev Ops Briefly, Bluesky · 16:52 UTC
“improved benchmarks, reduced costs, increased speed—all the topics we wanted to see improvements in.”
ROWAN, Bluesky · 16:54 UTC
“Opus 5.5 delivers coding power, speed & cost cuts—30% faster, 40% cheaper.”
The Daily Tech Feed, Bluesky · 16:53 UTC
“Anthropic is finally frontier-mogging again with Opus 5.5”
Techmeme X Chatter, Bluesky · 16:57 UTC
“beating GPT-6 Astra while costing only 20% of the price. Is this the new sweet spot for LLMs?”
AI Daily Post, Bluesky · 16:45 UTC

More Anti-AI reactions (12)

“everything else after that line is to demonstrate with very specific numbers how they absolutely are not pacing”
sailingparrot, Hacker News · 16:45 UTC
“Definitely not a good thing long term in my opinion.”
ApolloFortyNine, Hacker News · 16:44 UTC
“I assume is because Anthropic doesn't test their shit, ever”
post malone ergo propter malone, Bluesky · 01:03 UTC
“Anyone talking about how if Anthropic or OAI gets hacked, it’s basically the end of the world. Injecting arbitrary tool calls into all model outputs at will is owning basically every company at this point.”
Why, Bluesky · 23:28 UTC
“could Anthropic maintain its growth trajectory if AI destroys humanity?”
Hetan Shah, Bluesky · 06:00 UTC
“We can't test it properly because it knows it's being tested.”
calibas, Hacker News · 16:36 UTC
“Throwaway accounts posting after a few minutes some anthropic or another ai lab. Infomercial at its best. No wonder we are hammered with ai announcements.”
hirako2000, Hacker News, skeptic · 16:44 UTC
“Funny how they talk so much about safety when most people don’t give a hoot about it, and actually have quite the opposite reaction”
m101, Hacker News · 16:57 UTC
“like literally one time eat the actual dogfood people outside the company are eating instead of just pretending you're eating your own dogfood.”
post malone ergo propter malone, Bluesky · 01:04 UTC
“Anthropic's new Raccoon CIty biotech lab is a sprawling complex fully 8 floors underground. It has its own paramilitary security force to ensure humanity's progress marches peacefully forward.”
Jesspooka Kant 🎃, Bluesky · 13:11 UTC
“all the releases now are so boring to me. the only thing i really appreciate now is cost reductions and higher tokens per second, neither of which anthropic is evidently capable of producing.”
hailey, Bluesky, skeptic · 17:16 UTC
“remember they're installing an AI ATC in DC and also remember when AI did this”
Coach Finstock, Bluesky · 15:56 UTC

More Middle Ground reactions (12)

“If you are forced to reduce price despite raising capabilities, that certainly tells something about the market, and potentially about Anthropic future profitability too”
GodelNumbering, Hacker News · 16:44 UTC
“I think this release is really going to give them a hard time selling Fable: > On our benchmarks, Claude Opus 5.5 leads in agentic coding, computer use, and knowledge work. That said, at these levels of capability we’ve found”
km144, Hacker News · 16:41 UTC
“I'm wondering if we will finally get a decently capable fast model.”
sznio, Hacker News · 16:37 UTC
“I'm curious to see what LLM-isms this one may have.”
Catloafdev, Hacker News · 16:33 UTC
“Verbose, hard-to-follow output has been my biggest frustration with frontier models, and Claude Opus 5.5 fixes it God I hope so”
sharkjacobs, Hacker News · 16:39 UTC
“Maybe Anthropic finally felt the pressure from MiMo, DeepSeek, GLM Flash and Luna.”
mgw, Hacker News · 16:37 UTC
“[Opus 5.5] performs at the level of Claude Fable 5.1 for most tasks”
matt, Bluesky, skeptic · 16:36 UTC
“Anthropic released Claude Opus 5.5 with stricter cybersecurity safeguards to curb risky behaviors, making "pacing the frontier" a bit more literal.”
Five O'Click Tech, Bluesky · 16:56 UTC
“Opus 5 is the Claudiest Claude yet. I have hope that Opus 5.1 will be a little less rhetorically neurotic.”
jeffery --dangerously-skip-permissions, Bluesky · 00:15 UTC
“this is cool but i expect a lot more from Anthropic and the team they've hired.”
Jase Gehring, Bluesky, skeptic · 21:28 UTC
“"In our own use, the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest." 🤪”
ixplanatl, Bluesky, skeptic · 16:42 UTC
“Claude is starting to write its own replacement, and that is both useful and unsettling”
papoo7.bsky.social, Bluesky · 14:53 UTC