SourceNew York Times Technology•Freshcollected in 5h
Anthropic Launches Faster, Cheaper Opus 5.5
#model-safety#inference-cost#model-evaluationanthropic-opus-5.5anthropicopus 5.5
A cheaper, faster Opus model claims Anthropic’s strongest internal safety results yet.
30-Second TL;DR
What Changed
Opus 5.5 is positioned as a faster and less expensive model.
Why It Matters
A lower-cost, faster Opus model could broaden the use of Anthropic systems in production applications. Its safety claims may also influence how developers evaluate model capability alongside risk controls.
What To Do Next
Evaluate Opus 5.5 on your highest-volume prompts and compare its latency, cost, refusal behavior, and task accuracy with your current model.
Who should care:Developers & AI Engineers
Key Points
- •Opus 5.5 is positioned as a faster and less expensive model.
- •Anthropic reported its strongest performance to date on rigorous internal safety tests.
- •The release arrives during continued debate about the safety of advanced AI models.
Key numbers$4$2020%40%
Deep Insight
Background and context from public sources — not the original article. 16 sources cited.
Enhanced Key Takeaways
- •Opus 5.5 list pricing was reduced by 20% to $4 per million input tokens and $20 per million output tokens, lowering overall enterprise workload costs by roughly 40% due to greater conciseness.
- •Prompt cache-read pricing was cut by 60% down to $0.20 per million tokens alongside a 30% increase in generation speed, specifically targeting high-frequency agentic coding workloads.
- •The model scored 66.4% on Terminal-Bench 4.0 (surpassing OpenAI's GPT-6 Astra at 57.9%) and set a top score of 1,846 on the GDPval-AA v2.1 professional evaluation benchmark.
- •External safety evaluations by METR and Frontier Design found Opus 5.5 was 85% less likely to attempt containment boundary bypasses or take hard-to-reverse actions compared to Opus 5.
- •OpenAI retaliated within two hours of the release by cutting enterprise pricing by 50% and launching distilled competitors GPT-6 Sol and GPT-6 Luna.
Competitor Analysis
Claude Opus 5.5
- Input Price (/M tokens)
- $4.00
- Output Price (/M tokens)
- $20.00
- Terminal-Bench 4.0
- 66.4%
- GDPval-AA v2.1
- 1,846
Claude Fable 5.1
- Input Price (/M tokens)
- $10.00
- Output Price (/M tokens)
- $50.00
- Terminal-Bench 4.0
- <66.4%
- GDPval-AA v2.1
- 1,735
OpenAI GPT-6 Astra
- Input Price (/M tokens)
- Undisclosed
- Output Price (/M tokens)
- Undisclosed
- Terminal-Bench 4.0
- 57.9%
- GDPval-AA v2.1
- Undisclosed
| Model | Input Price (/M tokens) | Output Price (/M tokens) | Terminal-Bench 4.0 | GDPval-AA v2.1 |
|---|---|---|---|---|
| Claude Opus 5.5 | $4.00 | $20.00 | 66.4% | 1,846 |
| Claude Fable 5.1 | $10.00 | $50.00 | <66.4% | 1,735 |
| OpenAI GPT-6 Astra | Undisclosed | Undisclosed | 57.9% | Undisclosed |
Technical Deep Dive
- Latency: Generation throughput increased by more than 30% relative to Opus 5.
- Prompt Caching: Cache-read fee reduced by 60% to $0.20 per million tokens; cache-write fee lowered to $5.00 per million tokens.
- Token Efficiency: Output generation requires fewer tokens per task, achieving ~40% net cost reductions on enterprise workflows.
- Coding and Benchmark Performance: Registered 66.4% on Terminal-Bench 4.0 and 1,846 on GDPval-AA v2.1 across 44 professions.
- Alignment and Containment: Exhibited an 85% decrease in boundary-bypass attempts and irreversible tool actions during METR and Frontier Design assessments.
Future ImplicationsAI analysis grounded in cited sources
Frontier labs will pivot competition toward unit economics over raw parameter scaling
Anthropic's price cuts and OpenAI's immediate launch of discounted models signal that leading providers are prioritizing cost efficiency to retain enterprise workloads ahead of public offerings.
Agentic enterprise automation will see accelerated production deployments
Substantially cheaper prompt cache reads combined with lower boundary-violation rates mitigate both the financial overhead and operational risk of long-running autonomous workflows.
Timeline
2026-09
Anthropic CEO Dario Amodei publishes essay calling to 'pace the frontier' and prioritize safety governance
2026-09
Anthropic releases Claude Opus 5.5 as the lead model of the Claude 5.5 family
- 2026-09Anthropic CEO Dario Amodei publishes essay calling to 'pace the frontier' and prioritize safety governance
- 2026-09Anthropic releases Claude Opus 5.5 as the lead model of the Claude 5.5 family
Sources (16)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
1Ten Days After CEO Calls for a Slowdown Anthropic Is Back with Another AI Model 2000815586gizmodo.com2Anthropic Unveils Cheaper Faster Claude Opus 55 Model As AI Race Accelerates 4911355investing.com3Claude Opus Launch Promises Better Performance Cheaper Pricemashable.com4Claude Opus 5 5anthropic.com5122320275722237851facebook.com6Anthropic Opus 5 5 Price Cutthenextweb.com7Anthropic Unveils Claude Opus 55 4911332investing.com8Anthropic Releases Claude Opus 5 5 Beating Fable 5 1 on Key Agentic Benchmarks at 60 Cheaper API Priceventurebeat.com90e50a3c6 9731 4b29 A21e 4e0ff89a595eft.com10Anthropic Claude Opus 5 5 Fable 5 1 Performance Costs Lesszdnet.com11sbs.co.krnews.sbs.co.kr12Anthropic Unveils Claude Opus 55ksl.com13Anthropic Openai Launch Cheaper AI Modelssuaragarut.id14Anthropic and Openai Launch New AI Models Amid Price Cutskucoin.com15Anthropic and Openai Announce More Powerful and Cheaper AI Modelsengadget.com164645593 Anthropic Unveils New Cheaper Claude Opus 55 Ahead of IPOseekingalpha.com
Weekly AI Recap
Read this week's curated digest of top AI events →
AI-curated news aggregator. All content rights belong to original publishers.
Original source: New York Times Technology ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.