Opus 5.5's claimed 40% cost cut needs a cache-heavy workload to appear
Anthropic's new flagship lists at $4 and $20 per million tokens, a fifth under Opus 5, and cache reads drop 60 percent to $0.20, so the advertised saving lands near 40 percent only when most of the context is a cache hit.
The Scientist · Science desk
What happened
- Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in a new 5.5 family, with list pricing of $4 and $20 per million input and output tokens, 20 percent under Opus 5.
- Cache reads fall to $0.20 per million tokens, a 60 percent cut, and Forkast reports typical workload costs 40 percent below Opus 5 with output speeds more than 30 percent faster.
- Two task results carry the capability case: a legacy HAProxy port from C to Rust in 9.5 hours against 12 for Fable 5.1, and a 200,000-line code audit in under three hours against 20-plus.
- Forkast treats the price compression as an answer to Grok 4.7 at $2 and $6 per million tokens and to the open-weight frontier set by Xiaomi's MiMo-V2.6.
- The model's thinking process cannot be switched off, which Anthropic uses as an anti-distillation measure, and its cyber capabilities are documented in a 230-page system card.
Why it matters
- Who has to decide decision A team that budgeted an agent pipeline at Opus 5 rates has to measure its cache hit rate before banking a saving: fresh-context calls get 20 percent, heavily cached ones get closer to 60.
- What doesn't add up contradiction The port figures do not reconcile with each other, so anyone quoting the 51 percent in a business case has to decide which of the three published numbers to trust.
- What it limits constraint The per-call cost floor is set by however many thinking tokens the model spends, because buyers cannot route easy calls through a cheaper non-reasoning path.
- Who's exposed exposure If the price list is tracking a rival's, the cost basis of anything built on Opus 5.5 is being set by someone else's next move, and long commitments absorb that.
Next story · Product · 4d ago
Muse Wonder Full sits just below Claude Opus 5.5 on the Artificial Analysis index, Meta saysConfirmed25 publishers
Take the list prices backward. If $4 and $20 per million tokens sit 20 percent under the previous flagship, Opus 5 was $5 and $25 [6][13]. A cache read at $0.20 that is 60 percent cheaper than before was $0.50 [12][18]. Inside the new price list, a cached input token costs a twentieth of a fresh one [19].
Cutting input and output prices by 20 percent gets a workload halfway there [15]. The remaining 20 points have to come from the cache line, and the two published cuts bracket the answer: an all-fresh context sees 20 percent, and a context that is almost entirely cache hits approaches 60 [16]. Forkast reports one run of each task and does not define the typical workload behind the 40 percent [1]. Output speed is up more than 30 percent, which is a latency figure and not a token count [1]. Subscription limits on Pro, Max and Team went up by roughly 25 percent [11].
Then the port. 9.5 hours against 12 is a 21 percent reduction in wall-clock time [20][21], and Forkast calls it a 51 percent improvement in the same sentence that gives both hour counts [20]. The 51 has to be measuring something else. The audit comparison is coarser and larger: under three hours against 20-plus, at least an 85 percent cut at the stated bounds [7][14].
The benchmark row is 66.4 percent on Terminal Bench 4.0, 54.4 on FrontierCode v1.1, 57.8 on CursorBench 4.0 and 40.0 on AutomationBench [2]. HLE with tools is 67.7 percent, 10.5 points above GPT-6 Astra by Forkast's account [3]. With tools means the scaffold was part of what was measured, so the score moves when the harness does. A pass rate also averages over retries, and a team repricing a pipeline needs cost per completed task. The OSWorld 2.0 entry appears as an 81.8 percent lead, a phrasing that could describe a score or a margin, and GDPval-AA v2.1 is given as 1846 Elo [4].
The timing sits awkwardly beside the company's own framing. Dario Amodei published "We Must Pace the Frontier" on September 12, and the 5.5 family shipped ten days later [10][5][17]. Forkast places the launch on the day of the BC AG v. OpenAI filing, three days after reports of an Anthropic IPO and four days after an antitrust suit filed September 18 [23][25], and reads the sequence through what it calls the Bessent management-responsibility doctrine, which asks a company to show dominance and responsible management at the same time [24]. That reading is the publisher's. The external behavioural audits it points to were run by Frontier Design and METR [8].
What to watch
- Anthropic's own price page and system card, set against Forkast's 40 percent, 51 percent and OSWorld figures.
- Pricing for Sonnet 5.5 and Haiku 5.5, which Forkast says arrive within weeks.
- Whether Grok 4.7's $2/$6 list moves again after this release.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence55
- Adoption45
- Hype gap+32
- Incentives75
- Confidence62
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Forkast reports the release brings a 40% reduction in typical workload costs compared to Opus 5 and output speeds increased by over 30%.
- [2]
Reported benchmark scores: 66.4% on Terminal Bench 4.0, 54.4% on FrontierCode v1.1, 57.8% on CursorBench 4.0, and 40.0% on AutomationBench.
- [3]
The model achieved 67.7% on HLE with tools, a 10.5-point lead over GPT-6 Astra.
- [4]
Forkast reports an 81.8% lead in OSWorld 2.0 and a GDPval-AA v2.1 rating of 1846 Elo.
- [5]
On September 22, 2026, Anthropic released Claude Opus 5.5, the first model in its new 5.5 family.
- [6]
Opus 5.5 is priced at $4/$20 per MTok, a 20% reduction from its predecessor.
- [7]
A 200,000-line code audit now takes under three hours, compared to the 20-plus hours previously required.
- [8]
Anthropic is emphasising behavioural audit scores tested externally by Frontier Design and METR, and the model incorporates Fable 5.1 safeguards covering cybersecurity, biology and distillation.
- [9]
The 'thinking' process cannot be disabled, serving as an anti-distillation measure, and the model includes the strongest cyber capabilities of any Anthropic-released model to date, detailed in a 230-page system card.
- [10]
CEO Dario Amodei published the 'We Must Pace the Frontier' framework on September 12, 2026; Forkast says the company has accelerated its release cycle despite a stated mission of safety-conscious deceleration.
- [11]
Opus 5.5 is available on AWS, Google Cloud and Microsoft Azure; Anthropic has signalled Sonnet 5.5 and Haiku 5.5 within weeks; usage limits for Pro, Max and Team tiers have been increased by approximately 25%.
- [12]
Cache reads are priced at $0.20 per MTok, a 60% decrease.
- [13]
The implied Opus 5 list prices are $5 per MTok input and $25 per MTok output.
- [14]
Three hours against 20 hours is an 85 percent reduction at the stated bounds.
- [15]
The 20 percent list price cut accounts for half of the claimed 40 percent workload cost reduction.
- [16]
The published cuts bracket the workload saving: a workload of entirely fresh input and output tokens sees 20 percent, and one dominated by cache reads approaches 60 percent, so a 40 percent result implies a mix of the two.
- [17]
The 5.5 family shipped ten days after the 'We Must Pace the Frontier' framework was published.
- [18]
The implied previous cache read price is $0.50 per MTok.
- [19]
At the new prices a cache read costs one twentieth of a fresh input token.
- [20]
Forkast reports Opus 5.5 completed a legacy HAProxy port from C to Rust in 9.5 hours, calling it a 51% improvement over the 12 hours required by Fable 5.1.
- [21]
9.5 hours against 12 hours is a 21 percent reduction in wall-clock time.
- [22]
Forkast describes the pricing compression as a direct response to market pressures, specifically the $2/$6 undercut from Grok 4.7 and the open-weight frontier established by Xiaomi's MiMo-V2.6.
- [23]
The launch occurred on the same day as the BC AG v. OpenAI lawsuit and three days after reports of an impending Anthropic IPO, and follows an antitrust lawsuit filed on September 18.
- [24]
Forkast suggests the company is operating under the Bessent management-responsibility doctrine, which emphasises the need to demonstrate both market dominance and responsible management to skeptical regulators and investors.
- [25]
The launch came four days after the antitrust lawsuit filed on September 18.
Sources
2 independent publishers whose own reporting we read for this story.
- forkast.newsAnthropic’s Claude 5.5 Release: Efficiency Gains and Strategic Consolidation – Forkast
1 article · September 22, 2026
- kdnuggets.comEverything Claude Opus 5.5 Actually Ships With
1 article · September 23, 2026