Panel templates

GPT-5.5 interrogates Sonnet 4.6 across three rounds. Both thinking. For when you want the answer to come clean under pressure — or admit it can't.

Best for

Examine a hard question

Estimate

Quick · lower estimated cost

No description provided.

Best for

Improve a draft or plan

Estimate

Quick · lower estimated cost

Three on three. Each side gets one OpenAI, one Anthropic, one Chinese model. East vs West plus a tiebreaker. 70% threshold keeps the bar reasonable.

Best for

Compare two sides

Estimate

Balanced speed and cost

Four frontier experts, each with live web search. Round 1 is silent — no model sees another's answer. Round 2 reveals positions. We measure how much each model moved after seeing the others. For forecasting and multi-source research.

Best for

Get independent estimates

Estimate

Balanced speed and cost

Attacker (DeepSeek R1) breaks with three vectors. Defender (Claude Sonnet 5) patches. Judge (Kimi K2.6) scores. Two rounds, 90% threshold — the strictest red team on the shelf. For specs and RFCs.

Best for

Find risks and failure modes

Estimate

Quick · lower estimated cost

One on one. Claude Sonnet 5 argues for the motion. Grok 4.20 argues against. A chairman judges; the minority view is preserved. For binary decisions you keep putting off.

Best for

Compare two sides

Estimate

Quick · lower estimated cost

Sonnet 5 vs Grok 4.20, both thinking + web search. Each has to articulate the other's strongest case. 75% threshold, two rounds.

Best for

Resolve polarized views

Estimate

Quick · lower estimated cost

Two heavy-hitters spend two rounds tearing into each other. Quick, brutal, and surprisingly productive for thesis statements and policy drafts.

Best for

Challenge assumptions

Estimate

Quick · lower estimated cost

Five deep models, all thinking, 85% threshold. The most rigorous forecasting on the shelf. Three rounds.

Best for

Get independent estimates

Estimate

Balanced speed and cost

Three sharp models, all with thinking, 80% threshold. For when you need the assistant to push back hard on a draft you love.

Best for

Challenge assumptions

Estimate

Balanced speed and cost

Two on two. Proposition: GPT-5.4 + Gemini 3.1 Pro. Opposition: Claude Opus 4.8 + DeepSeek V4 Pro. Two rounds of opening and closing. The chairman declares the winner — both sides go on the record.

Best for

Compare two sides

Estimate

Balanced speed and cost

One on one. Gemini 3.1 Pro vs Llama 4 Maverick, both with web search. 60% threshold — the bar is low, but every claim needs a citation.

Best for

Compare two sides

Estimate

Quick · lower estimated cost

Three mixed models — one with thinking, one with web search, one baseline. Two rounds, 70% threshold. The generalist option for "what's likely to happen".

Best for

Get independent estimates

Estimate

Quick · lower estimated cost

Four different model families, no flags, 75% threshold. A balanced default for general-purpose writing and analysis.

Best for

Improve a draft or plan

Estimate

Balanced speed and cost

Four diverse models, two rounds, 70% threshold. For multi-perspective steel-manning where you want to see the same claim from four angles.

Best for

Resolve polarized views

Estimate

Balanced speed and cost

Four top-tier models with extended thinking. For the questions where 90% consensus is the only answer you trust under your own name.

Best for

Improve a draft or plan

Estimate

Balanced speed and cost

Three diverse models, no flags, 65% threshold. The lower bar forces real disagreement — perfect for stress-testing your own draft.

Best for

Challenge assumptions

Estimate

Balanced speed and cost

Three frontier models across five long rounds. The slow, methodical answer — for security audits, regulatory briefs, and any artifact that will be read in court.

Best for

Improve a draft or plan

Estimate

Balanced speed and cost