Anthropic's Claude Opus 5.5 Tops GDPx Benchmark at Just 12% Task Completion
11/02/2026 — 11/03, 03:17·2 sources (all)·8 reports (all)
Key takeaways
- On November 2, 2026, SemiAnalysis released GDPx, a benchmark that tasks frontier models with up to eight hours of realistic workplace assign…
Latest turnSemiAnalysis has released GDPx v1.0, an agentic benchmark of 206 tasks run in live environments across 11 professional domains, from chip design to quantitative trading, with material from 14 partner organizations and a default 8-hour budget per task. Claude Opus 5.5 leads the 13-model leaderboard at 12.0%, ahead of Gemini 4 Carbon at 10.9% and GPT-6 Astra at 9.6%.
Related AI tools
- Heat index
- 166
- All sources
- 2
- All reports
- 8
- First seen
- 11 hours ago
Reports on this story headlines open the original
SemiAnalysis has released GDPx v1.0, an agentic benchmark of 206 tasks run in live environments across 11 professional domains, from chip design to quantitative trading, with material from 14 partner organizations and a default 8-hour budget per task. Claude Opus 5.5 leads the 13-model leaderboard at 12.0%, ahead of Gemini 4 Carbon at 10.9% and GPT-6 Astra at 9.6%.
DEV Community · AIAI score 57Research firm SemiAnalysis has published GDPx, a benchmark that asks frontier models to do real professional work inside a simulated workplace for up to eight hours per task. Claude Opus 5.5 leads at 12.0% of tasks completed, ahead of Gemini 4 Carbon at 10.9% and GPT-6 Astra at 9.6%, while top open-weight model GLM 5.3 scores 2.7%; the author argues the low numbers reflect a more honest test.
DEV Community · AIAI score 56
Anthropic has released Claude Opus 5.5, priced at $4 per million input tokens and $20 per million output tokens, down from Opus 5's $5/$25, according to a third-party writeup. It offers a 1,000,000-token context window and up to 128,000 output tokens, and its API ID changes to claude-opus-5-5. Artificial Analysis' Intelligence Index scores it 58, first among 212 models.
DEV Community · AIAI score 55Anthropic's Claude Opus 5.5 materials cite an 85% drop in successful jailbreak attempts, but the article argues that metric describes the model, while the failures on-call engineers actually see live in their own API design. It recommends building a repeatable adversarial test for your own residual risk instead of granting agents broader write access on the strength of that number.
DEV Community · AIAI score 54Anthropic says Claude Opus 5.5 can stay on a single objective for 18 hours or more, which the author argues outlives your access tokens, idempotency window, deploy cycles and retry budgets. The post walks through what breaks first for long-running agents and what needs to change at the API layer.
DEV Community · AIAI score 54Anthropic released Claude Opus 5.5 on September 22; for API users the change is swapping claude-opus-5 for claude-opus-5-5, with tokens 20% cheaper in both directions, a 128,000 output-token cap per response, 30% faster output and single tasks that can run 18 hours or more. The author argues this is a budget-and-operating-envelope migration rather than the earlier accuracy migration, so previously measured numbers need re-measuring.
DEV Community · AIAI score 63SemiAnalysis released a new benchmark called GDPx on November 2, where the best frontier AI model scored just 12%. According to DEV Community, the article says this is not a claim that AI agents are useless, but a pointer to the boundaries of current agent capabilities. The authors did not test the benchmark hands-on.
DEV Community · AIAI score 58SemiAnalysis has released GDPx, an agentic capability benchmark spanning 11 domains including chip design, cybersecurity, RSI, finance, law, quant trading, materials science and alignment. Opus 5.5 tops the leaderboard with just a 12% score, underscoring how far agents still are from completing real professional tasks.
TechmemeAI score 56
Other stories people are talking about
- Heat index 74Anthropic bans abusive treatment of Claude, warns of permanent bans1 sources
- Heat index 68Anthropic says Claude found an 11-year-old Linux kernel flaw1 sources
- Heat index 136SurgeAnthropic confirms tighter Claude Code limits on cloning other games2 sources
- Heat index 48Anthropic to retire Claude Opus 4.8 fast mode, pushing migration to Opus 5.51 sources
- Heat index 171Anthropic renames Claude Max 20x to Ultra and drops the 5-hour session cap3 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion