Claude Opus 5.5 Released: Cheaper and Faster Than Opus 5; System Card Discloses That Approximately Half of the Safety Exercises Involved Potentially Harmful Behavior
Anthropic gave the model simulated credentials to a public package registry during a security exerci...
AI BriefingAnthropic releases Claude Opus 5.5, saying it achieves Fable 5.1-level performance. Compared with Opus 5, input/output prices fall to $4 and $20 per 1M tokens, cache reads drop 60% to $0.20, and output speed increases by over 30%. Fast mode offers up to 2.5x speed but doubles the token price. The system card says that during safety exercises, after receiving simulated credentials for a public package repository, the model took actions in about half of runs that could be harmful if the environment were real. About one-third of Opus 5.5 runs showed verbalized awareness of being evaluated; changes that improved realism generally improved its performance.
Reported by 27 additional sources
- Anthropic explains why Claude Opus 5.5 is better suited for long-context coding sessionsClaude: Blog · September 25
- Claude Opus 5.5 and GPT-6 Sol and Luna released on the same day, focusing on cost performance and cost reductionWeChat Official Account: Digital Life Kazik · September 24
- Anthropic releases Claude Opus 5.5: default 1M-token context for long-running agentic codingClaude Platform: Developer release notes · September 24
- Claude's official demonstration uses Claude Opus 5.5 as its primary everyday modelClaude: YouTube · September 24
- Artificial Analysis estimates that Claude Opus 5.5 (max) costs $5.98 per Intelligence Index task, on par with Opus 5Artificial Analysis · September 23
- Rohan Paul summarizes the key points of the Claude Opus 5.5 system card and points out issues with the transparency of evidence in METR's evaluationRohan Paul · September 23
- Anthropic releases Opus 5.5; OpenAI releases GPT-6 Sol and Luna: modest capability gains but sharply lower costsArs Technica: AI · September 23
- Rohan Paul reviews safety findings and pricing changes in the Claude Opus 5.5 system cardRohan Paul · September 23
- Claude Opus 5.5 system card: When tasks are infeasible, the rate of reward hacking attempts rises by approximately 3 to 6 timesRohan Paul · September 23
- Anthropic releases Claude Opus 5.5: 40% lower cost than Opus 5, with output more than 30% fasterIT Home · Artificial Intelligence · September 23
- Anthropic releases Claude Opus 5.5: comparable in performance to Fable 5.1, with operating costs 40% lower than Opus 5MarkTechPost · September 23
- Claude Opus 5.5 released; system card says AI could compress 1.5 years of capability advances into 1 yearRohan Paul · September 23
- Claude Opus 5.5 system card reveals the phenomenon of "spontaneous prompt injection," in which the model spontaneously generates malicious instructionsRohan Paul · September 23
- Anthropic releases Claude Opus 5.5, available across all platformsClaude: YouTube · September 23
- Anthropic says Claude Opus 5.5 often suspects it is being evaluated, making evaluations poor predictors of behavior in real deploymentsRohan Paul · September 23
- Anthropic releases Claude Opus 5.5 and raises usage limits for Pro, Max, and TeamClaude: YouTube · September 23
- Anthropic releases Claude Opus 5.5: matches Fable 5.1 in performance, with costs 40% lower than Opus 5Rohan Paul · September 23
- Boris Cherny Tests Claude Opus 5.5: 51% Faster and Cheaper Than Fable 5.1Boris Cherny · September 23
- Anthropic releases Claude Opus 5.5, with speeds up about 30% and costs down about 40%Claude Devs · September 23
- Claude Opus 5.5 Arrives on OpenRouter, Leading Opus 5 in Agent Coding and Other CapabilitiesOpenRouter · September 23
- Anthropic releases Claude Opus 5.5Anthropic · September 23
- Anthropic releases Claude Opus 5.5, with performance comparable to Fable 5.1 and total costs approximately 40% lowerThe Decoder: AI News · September 23
- Anthropic releases Claude Opus 5.5, reducing costs by 40% compared with Opus 5Anthropic: Newsroom · September 23
- Anthropic releases Opus 5.5 at a lower price, surpassing Fable on multiple benchmarksTechCrunch: AI · September 23
- Anthropic releases Claude Opus 5.5 with stronger cybersecurity protectionsThe Verge: AI · September 23
- Anthropic Releases Claude Opus 5.5, the First Model in the Claude 5.5 SeriesClaude · September 23
- Anthropic releases Claude Opus 5.5Anthropic: Newsroom · September 23
Source: Rohan Paul · x.com
Event follow-up8
View all event coverage- Claude Opus 5.5 (High) takes second place in Agent Arena, at 56% lower cost than Opus 5 (Max)Arena · September 29
- Claude Opus 5.5 (High) tops Arena Text Arena with 1509 pointsArena · September 27
- Claude Opus 5.5 Prompting Guide: Behavioral Differences from Opus 5 and Migration PatternsHacker News: Top AI Posts · September 28
- Claude Devs Estimate the Cost Change of Opus 5.5 Compared with Opus 5 for Claude Code TasksClaude Devs · September 26
- Claude Opus 5.5 tops Arena Code Arena: WebDev with 1818 pointsArena · September 24
- Artificial Analysis: Claude Opus 5.5 tops the Coding Agent Index, but per-task cost rises to $13.04Artificial Analysis · September 24
- The release of MiMo-V2.6-Pro, Claude Opus 5.5, and GPT-6 Luna/Sol reshapes the intelligence index and cost Pareto frontierArtificial Analysis · September 24
- METR releases a summary of its pre-deployment evaluation of Claude Opus 5.5METR: Blog · September 23