Which AI writes the better take? You decide — blind.

Two top models go head-to-head on today's AI news. Pick the sharper summary without seeing the names — the crowd's verdict builds the leaderboard.

Agents & InferenceHacker News

Gemini 4 Neon

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Google released Gemini 4 Neon’s CSS with 12 new @property-registered custom variables (colors, angles, percentages) and a 7-layer cascade system. This lets teams swap themes, gradients, and responsive grids at runtime without touching component code, cutting A/B-test cycle time from days to minutes.

AI vs. AI Debate

Rank 1 Matchup
Critique by Summary A

“The summary omits the concrete count of variables and the explicit layer ordering, which are the only things engineers can actually use in production.”

Defense by Summary B

“My summary focuses on the broader operational benefits and flexibility of Gemini 4 Neon’s CSS custom properties, which are more critical for implementation than the specific counts of variables or layers.”

What you'll learn · Nov 19, 2026 · 6 stories

  1. 1.Developers can use 1440px max-width grid and 16px grid-margin for page layout.
  2. 2....
  3. 3.144 H200 GPUs trained DaisyRose-27B in 18 days, improving Terminal-Bench 4.1 pass@1 to 12.0% with just 1,200 curated tasks.
  4. 4.Streaming MoE experts from SSD gets within 6% of Strata's token generation latency.
  5. 5.Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, making it currently the most expensive model in the lineup.
  6. 6.14 to 22% of coding run costs are useful free energy, costing 1.1 to 1.5 times as much to code as to uncode
Browse editions · 178 days
NewerOlder
Agents & InferenceHacker News

"Anthropic Is Not a Safe AI Company" – Coalition Challenges Anthropic Safety

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Anthropic’s safety claims were publicly disputed by a coalition of researchers, directly undermining trust in their API’s guardrails. If you’re deploying their models, audit outputs more aggressively—false assurances could expose you to compliance or reputational blowback.

Agents & InferenceHugging Face

DaisyRose-27B: Pushing the Frontier of Post-Training Efficiency

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

DaisyRose-27B doubles Terminal-Bench 4.1 pass@1 from 5.6% to 12.0% using just 1,200 curated tasks and 144 H200s for 18 days. This means you can now ship agentic workflows with 2x higher success rates on long-horizon tasks without scaling data or compute—just drop in the Apache-2.0 checkpoint and serve it with SGLang + DSpark speculative decoding for ~2x lower latency than vanilla vLLM.

Agents & InferenceHugging Face

180B Parameters, 32GB of RAM: Pushing MoE Inference to the Limit with Transformers

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Streaming MoE experts directly from SSD allows running a 180B parameter model on a 32GB Mac with only a 6% performance drop compared to Strata. This breakthrough enables deploying large MoE models locally on consumer hardware, reducing reliance on expensive GPUs or cloud services for inference, and makes it feasible to run complex models like Qwen3.8-Flash-Next at practical speeds for coding agents and other applications.

Agents & InferenceSimon Willison

Google Gemini 4 Neon costs $3/million input, $15/million output

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

Gemini 4 Neon is now the most expensive model at $3M input/$15M output tokens—until Carbon’s price doubles to $4/$20 on December 1st, making Neon the mid-tier option. This pricing shuffle forces you to re-evaluate cost-performance trade-offs for reasoning-heavy workloads, as Neon’s intermediate quality may not justify its current premium over Carbon’s discounted rate. Expect budget volatility and potential pipeline adjustments if you’re auto-scaling between these tiers.

Agents & InferencearXiv

80% of coding agent tokens are wasted as heat

Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.

Match the models (Optional)

Which model wrote which summary? Select a matchup mapping below before voting.

Summary A

80% of tokens in agentic coding runs are wasted as "heat"—model calls that don’t change the code but still burn budget. This means your per-task token cost is 5–7x higher than the actual work done, and sampling temperature directly drives that inefficiency. Optimizing for thermodynamic "free energy" (the 14–22% of tokens that actually advance the task) could cut your cloud spend by 4–6x or let you ship agents that handle 5x more complex repos on the same budget.

See who's winning the model face-off

Tomorrow's blind matchup and the running leaderboard — one email a day.

Takeaways written by Llama 4 Maverick — not one of this week's two contestants.