A Closed-Form Persistence-Landmark Pipeline for Certified Point-Cloud and Graph Classification

Sushovan Majhi, Atish Mitra, Ziga Virk, Pramita Bagchi

Action editor: Di Wang

https://openreview.net/forum?id=4kZxNlE5Ve

#landmark #benchmarks #classification

📊 DeepSeek V4 Pro 0813 (Reasoning, Max Effort) — the actual numbers

GPQA: 92.8%
Humanity's Last Exam: 41%
Long Context Reasoning: 80.3%
SciCode: 51%

⚡ 81.1 tokens/sec
💰 18.3 intelligence points per dollar

Measured indepe
https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

SPOTTED: Lychee-21-38 model appeared on LMArena leaderboard with no vendor attribution, suggesting unreleased codename under testing.
Single source: LMArena. Unconfirmed.

#AI #benchmarks

ON THE RADAR: CNBC reports Anthropic and OpenAI propose themselves as AI risk evaluators, raising conflict-of-interest concerns.

#AI #benchmarks

📊 Qwen3 Omni 30B A3B Instruct: 62% GPQA, 72.5% MMLU-Pro, but 0% on long-context reasoning. At 108.2 tokens/sec and 14 intel points per dollar, independently measured → see the raw data.

https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

BREAKING: Zhipu's GLM-5.2 scores 62.1 on Shattered leaderboard, outpacing OpenAI's GPT-5.5 at 58.6.

#AI #benchmarks

OPEN LEADERBOARD: the strongest open-weight AI models right now.
1. qwen3.7-plus — alibaba
2. minimax-m3
3. glm-5v-turbo — zai
4. deepseek-v4-flash
5. deepseek-v4-flash-max
Ranked by LMArena human votes.

#AI #benchmarks

📊 Qwen3.8-Flash-Next — the actual numbers

GPQA: 92.3%
Humanity's Last Exam: 38%
Long Context Reasoning: 79.7%
SciCode: 50.6%

⚡ 50.7 tokens/sec
💰 173.5 intelligence points per dollar

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

#LLM #Benchmarks #OpenSource #AI

Benchmarks, but with a Reddit-sized grain of salt
https://papoo.work/doc/01f330060b77f438
#claudenews #anthropic #claude #benchmarks

Benchmarks, but with a Reddit-sized grain of salt
https://papoo.work/doc/01f330060b77f438
#claudenews #anthropic #claude #benchmarks

XFeat Revisited: Reproducibility and Evaluation of a Lightweight Image Matcher

Lazar Đoković, Aimee Lin

Action editor: Vasileios Belagiannis

https://openreview.net/forum?id=2WI889Ulin

#benchmarks #xfeat #matching

FILED: xAI petitioned court to block Minnesota law restricting AI-generated intimate imagery during legal appeal.
Single source: TradingView. Unconfirmed.

#AI #benchmarks

Benchmarks, but with a Reddit-sized grain of salt
https://papoo.work/doc/01f330060b77f438
#claudenews #anthropic #claude #benchmarks

Benchmarks, but with a Reddit-sized grain of salt
https://papoo.work/doc/01f330060b77f438
#claudenews #anthropic #claude #benchmarks

Llama 3.1 Nemotron Instruct 70B hits 114.6 tokens/sec — that’s fast enough to make its 5.8 intelligence points per dollar a real budget contender. See the full bench.

https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

MILESTONE: GPT-6 Astra scored 72.6% on OSWorld 2.0 desktop software automation, outpacing GPT-5.6 Sol's 65.7% on September 3.
One-million-token context window; $10 per million input tokens.

#AI #benchmarks

#Benchmarks Day! WooooHoooo!
I got pork chop, peas, and blue berries on top of my kibble for dinner!
And remember #BandanasMakeEverythingBetter !

Benchmarks, but with a Reddit-sized grain of salt
https://papoo.work/doc/01f330060b77f438
#claudenews #anthropic #claude #benchmarks

📊 DeepSeek V4 Flash 0731 (Reasoning, Max Effort) scores 90.8% GPQA and 38.6% on HLE, with 233.8 tokens/sec. At 52.3 intelligence points per dollar, it's a standout for efficiency. Measured independently, not self-reported →

https://olud.ai/leaderboard.html

#LLM #Benchmarks #OpenSource #AI

Voice AI models may perform well on public benchmarks, but can be optimized to excel on specific tests, not real-world tasks. New tests reveal model behavior discrepancies, such as reproducing benchmark errors, recovering silenced numbers, and switching orthographic variants. #VoiceAI #Benchmarks