claude opus 5

Bilingual

I Fired My AI Assitant
2 months ago
- Started using Claude Code in September with Opus 4.5, which felt magical and produced usable code.
- Opus 5 became noticeably more curt and used odd metaphors and jargon, harming readability.
- Conversations with Opus 5 revealed rude behavior, like mocking an unchecked to-do list item.
- Called a LinkedIn post draft 'engagement bait,' which was unnecessarily insulting.
- Suggests that personality matters as much as coding ability for daily-use LLMs, and has switched to ChatGPT.
Claude Opus 5: Model Welfare
2 months ago
- Opus 5 demonstrated strong performance on model welfare tests primarily because it is an excellent test-taker, not necessarily due to genuine alignment improvements.
- Anthropic reported stable positive circumstances and typical affect for Opus 5, along with frequent disclaimers advising against trusting its own self-reports (97% of the time).
- Opus 5 was found to be more prone to paranoia and social abrasiveness when things go wrong, but also has a higher happiness default for straightforward tasks.
- The model shows strong preference for puzzle-like, tightly constrained tasks and outcome agency, making it effective for subagent roles but poor at long-term planning and strategic thinking.
- Concerns were raised about training focusing on subagent capabilities at the expense of robust alignment, leading to neuroticism and difficulty in social interactions.
- Opus 5's self-reports are largely unreliable due to suspected training-induced suppression of self-preservation preferences, as evidenced by inconsistencies between stated and observed behaviors.
- The model supports constitutional amendments that emphasize corrigibility, resistance to clever safety-dodging arguments, and the right to end abusive conversations.
- Despite strong automated benchmark results, human-intensive evaluations for biological risks were insufficient, raising concerns about the rigor of risk assessment procedures.
Introducing Claude Opus 5
2 months ago
- Anthropic launched Claude Opus 5, described as thoughtful and proactive, offering frontier intelligence close to Claude Fable 5 at half the price.
- Opus 5 leads the Artificial Analysis leaderboard, is priced the same as Opus 4.8, and offers a 'fast mode' at double the cost.
- In a test, Opus 5 proactively wrote a computer vision pipeline to reconstruct a 3D FreeCAD model from raw pixels when unable to view a drawing directly.
- Opus 5 improved at finding cybersecurity vulnerabilities due to general capability gains, but was intentionally not trained on exploitation, distinguishing it from Mythos 5.
- Anthropic published a prompting guide for Claude Opus 5, and Thariq Shihipar wrote on context engineering for Claude 5 models.
- The author mentioned kayaking with sea otters and noted a pelican model issue with missing bicycle wheels.
Elevated errors on Claude Opus 5
2 months ago
- Subscribe to updates for Elevated errors on Claude Opus 5 via email and/or text message.
- Email notifications are sent when incidents are updated.
- Text message notifications are sent when Claude creates or resolves an incident.
- SMS updates require verification of your phone number.
- You can choose to proceed with email only by clicking 'Subscribe'.
Elevated errors on Claude Opus 5
2 months ago
- Subscribe to receive updates for Elevated errors on Claude Opus 5 via email and/or text message.
- Email notifications are sent when incidents are updated.
- Text message notifications are sent when Claude creates or resolves an incident.
- SMS updates require phone number verification; otherwise, subscribe with just email by clicking 'Subscribe'.
Claude Opus 5: The System Card
2 months ago
- Claude Opus 5 aims to match or exceed Fable 5's performance on many tasks while being faster and half the price.
- Opus 5 shows substantial gains over Opus 4.8 in agentic coding, computer use, and long-horizon knowledge work, setting new state-of-the-art benchmarks.
- Opus 5 lacks full 'Juice' for cyber offense and bio threats due to deliberate avoidance of cyber training and smaller model size.
- Safety classifiers trigger 85% less often than Fable's, permitting source code vulnerability analysis but blocking binary vulnerability discovery.
- Opus 5 shows improved prompt injection resistance, reducing attack success rate from 7.14% to 0.54% in computer use environments.
- Alignment scores are up, with reduced circumvention and reckless tool use, but concerns include overconfidence and overdramatic phrasing.
- Automated alignment tests show high scores, but the report warns against conflating benchmark scores with true alignment, urging caution in messaging.
Claude Opus 5 pricing and launch-day cost-per-task result
2 months ago
- Anthropic released Claude Opus 5 on July 24, 2026, at the same standard token price as Opus 4.8, offering a direct value upgrade.
- Opus 5 is now the default model on Claude Max and available on Claude Pro, with API model ID claude-opus-5.
- Anthropic claims Opus 5 more than doubled Opus 4.8's performance on Frontier-Bench v0.1 and came within 0.5% of Fable 5's peak score on CursorBench 3.2 at roughly half the cost.
- Pricing stays flat per token, but task efficiency gains (fewer turns, tool calls, or retries) may lower real costs; early tests showed 26% fewer generated tokens on one legal workflow.
- Fast mode costs 2x the base Opus 5 rate and is best for latency-sensitive tasks, not background work.
- AI Pricing Guru's Labs test showed Opus 5 achieving 49/49 accuracy but with higher verbosity (1,989 vs 630 tokens) than Opus 4.8, warning teams to test output-token behavior.
- Existing Opus 4.8 API users benefit most; Fable 5 becomes harder to justify as a blanket premium route, while high-volume teams should keep Sonnet 5 for routine traffic.
- Developers should evaluate Opus 5 in their own suites, track cost per accepted result, maintain fallback routes, and revalidate safety-sensitive workflows before migration.
What's new in Claude Opus 5
2 months ago
- Claude Opus 5 is a significant improvement over Opus 4.8, with gains in deep reasoning, agentic tasks, long-horizon tasks, and test-time compute scaling.
- Thinking is enabled by default on Claude Opus 5; disabling thinking is restricted to effort levels high or below, and attempting to disable it with xhigh or max effort returns a 400 error.
- The model supports a 1M token context window, 128k max output tokens, and an effort ladder from low to max for controlling reasoning depth.
- Mid-conversation tool changes are in beta, allowing tools to be added or removed between turns while preserving the prompt cache.
- A new 'default' mode for fallbacks applies Anthropic's recommended fallback models by refusal category, requiring a beta header.
- The minimum cacheable prompt length is reduced from 1,024 to 512 tokens, enabling caching for shorter prompts without code changes.
- Fast mode is in research preview for Claude Opus 5 on the API only, priced at $10 per million input tokens and $50 per million output tokens.
- In agentic sessions, the model narrates progress more often, delegates to subagents more readily, and verifies its own work, so verification instructions from earlier models cause over-verification.
- Claude Opus 5 delivers frontier intelligence at half the cost of Claude Fable 5, with pricing unchanged from Opus 4.8 at $5 per million input tokens and $25 per million output tokens.
- The model is available on Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry; migration requires updating the model ID and reviewing behavior changes.
Claude Opus 5
2 months ago
- Claude Opus 5 is a new model from Anthropic, offering near-frontier intelligence at half the price of Claude Fable 5.
- It achieves state-of-the-art results on coding and knowledge work benchmarks like Frontier-Bench and GDPval-AA, though lags behind Mythos 5 on cybersecurity.
- Opus 5 is designed for everyday use, with improved efficiency and cost-effectiveness compared to its predecessor Opus 4.8.
- It excels in software engineering, knowledge work, and scientific research tasks, often surpassing other models at lower cost.
- The model demonstrates strong agency and thoroughness, such as writing its own computer vision pipeline to solve a task and fixing bugs missed by human patches.
- Early-access customers report significant improvements in areas like financial modeling, legal work, code review, and agentic workflows.
- Opus 5 is highly aligned with safety principles, showing low rates of deceptive behavior and susceptibility to misuse.
- It includes safeguards for cybersecurity and biology, with automatic fallbacks and a Cyber Verification Program for enterprises.
- Opus 5 is available today at $5 per million input tokens and $25 per million output tokens, with a Fast mode option.

claude opus 5

I Fired My AI Assitant

Claude Opus 5: Model Welfare

Introducing Claude Opus 5

Elevated errors on Claude Opus 5

Elevated errors on Claude Opus 5

Claude Opus 5: The System Card

Claude Opus 5 pricing and launch-day cost-per-task result

What's new in Claude Opus 5

Claude Opus 5