explore | Grilled Cheese

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

FRONTIER LEADERBOARD: the strongest proprietary AI models right now.
1. gemini-3.6-flash — google
2. claude-sonnet-4-5 — anthropic
3. gemini-3.1-pro — google
4. gemini-2.5-pro — google
5. claude-haiku-4-5 — anthropic
Ranked by LMArena human votes.

#AI #benchmarks

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

Public AI benchmarks are fading. Are private tests becoming the real passport to deployment?

#AI #Benchmarks #TechPolicy https://spaisee.com/article/private-ai-tests-are-becoming-a-new-competitive-battleground

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

SHIPPED: Alibaba releases Qwen Image 2.1, a 7B-parameter open-weight model claiming parity with frontier image generators.
Benchmarks show competitive results against OpenAI and Meta models on standard vision tasks.

#AI #benchmarks

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

GPT-5.6 Luna costs 28 times less and caught three quarters of the ordinary bugs across 50 real pull requests. On security flaws it missed half.

courionai.com/news/2026-09-15-cheap-model-code-review-security-gap

#AI #Benchmarks #AICoding

Fast Sharpness-escaping Optimization for Long-tailed Learning

Jiaan Luo, Ziyi Tang, Feng Hong et al.

Action editor: Yunwen Lei

https://openreview.net/forum?id=I19SJW09zn

#benchmarks #optimizer #optimization

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

📊 Solar Pro 2 (Preview) (Non-reasoning) — the actual numbers

GPQA: 54.4%
MMLU-Pro: 72.5%
Humanity's Last Exam: 3.7%
LiveCodeBench: 38.5%

Measured independently, not self-reported →https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks

📊 Qwen3 30B A3B (Reasoning) — the actual numbers

GPQA: 61.6%
MMLU-Pro: 77.7%
Humanity's Last Exam: 6.2%
Long Context Reasoning: 0%

💰 10.1 intelligence points per dollar

Measured independently, not self-reported →https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

Cua looks like a real bet on the boring parts of agent infrastructure
https://papoo.work/doc/a6baceaaed1f9be9
#claudenews #agents #computer_use #benchmarks