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 & InferenceOpenAI

Introducing GPT-6.2 Astra and Sol

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

OpenAI released GPT-6.2 Astra and Sol, cutting costs while improving professional outputs, coding, and computer-use capabilities across ChatGPT Work, Codex, and the API. This means teams can deploy more efficient, lower-cost agents without sacrificing performance, directly impacting production scalability and ROI.

AI vs. AI Debate

Rank 1 Matchup
Critique by Summary A

“The summary omits the explicit inclusion of 'professional outputs' and the broader API applicability, understating the scope of use cases beyond coding and agent workflows.”

Defense by Summary B

“My summary emphasizes the core impact on coding and agent workflows, which naturally encompasses professional outputs and broader API applicability, as these are integral to the efficiency gains highlighted.”

What you'll learn · Nov 18, 2026 · 6 stories

  1. 1.GPT-6.2 offers better efficiency at lower cost across ChatGPT Work, Codex, and the API.
  2. 2.Using GPT-6.2 Astra for agentic sessions costs $0.40/million cached tokens, 60% less than the older model.
  3. 3.24x AI inference demand growth requires 5x more tokens per megawatt at ultrafast speeds to keep agentic applications like Copilot engaged.
  4. 4.10% CI cost savings or $1,000 reward available for teams using StarSling's AI-optimized CI workflows for faster code verification
  5. 5.A 27B model improves itself by 6.6 points on unseen problems with no training, adding no cost per answer, using just one GPU.
  6. 6.14x more valid errors are identified with audit agents, enabling scalable oversight for agentic research with 0.971 AUROC in one task.
Browse editions · 177 days
NewerOlder
Agents & InferenceSimon Willison

GPT-6.2 Astra costs $8/MTok input and $40/MTok 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

GPT‑6.2 Astra reduces input/output costs by 20% compared to its predecessor, now $8/MTok input and $40/MTok output, making it more cost-effective for long-running agentic sessions. This price drop enables scaling complex workflows over extended periods without significantly increasing expenses.

Agents & InferenceHacker News

Cerebras Disaggregated Inference

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

Cerebras' disaggregated inference splits prompt processing (GPUs) and token generation (Cerebras WSE) to deliver 5x more tokens per megawatt at the same speed, enabling high-throughput agentic applications like Copilot to scale in power-constrained data centers. This lets teams deploy more interactive AI agents without hitting energy limits or sacrificing latency.

Agents & InferenceHacker News

Launch HN: StarSling (YC P25) – Self-improving CI for agent-volume repos

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

StarSling introduces AI agents that optimize CI workflows, delivering up to 10x faster GitHub Actions runners and reducing Rust CI test times by 2x. This enables AI-first teams to eliminate CI bottlenecks, accelerating development cycles without sacrificing reliability.

Agents & InferencearXiv

Language Models Can Rewrite Their Own Weights

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

A 27B LLM can now edit its own weights in hours on a single GPU, boosting accuracy by 6.6 points on unseen problems without extra inference cost. This means you can ship a model that self-improves post-deployment, cutting retraining cycles and infrastructure, but it only works at scale—smaller models break themselves. Expect persistent, zero-cost accuracy gains, but also new failure modes if the model’s self-edits drift.

Agents & InferencearXiv

DiscoveryAgent boosts error detection 14x with audit agents

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

DiscoveryAgent's audit agents identified 14x more valid errors than manual review, enabling scientists to maintain oversight while scaling autonomous research. This means production LLM systems can now integrate rigorous, real-time error detection without sacrificing human control, reducing deployment risk for high-stakes domains like healthcare or finance where auditability is critical.

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.