Official run

Claude Sonnet 5.5High

In this Twenty Questions LLM benchmark, Claude Sonnet 5.5 (high) ranks 8 of 18 with a 17.83 question score and 90% success. Lower is better. The run contains 30 scored episodes across 10 subjects with 3 trials per subject. Choose a subject to inspect its attempts.

Rank 8 of 18anthropicHigh

Question score17.83questions · lower is better95% CI 15.49–20.18

Success: 90%
Contract compliance: 100%
Guesser cost: $1.38
Guesser time: 27m 34s

Drill down

Subjects.

01Albert Einstein10.33 avg 02Albert Schweitzer20.67 avg 03Garfield11.33 avg 04Achilles9.33 avg 05Genghis Khan9.33 avg 06Bike pump32.67 avg 07Spider web28.00 avg 08Eyebrow15.00 avg 09Moon9.67 avg 10Door handle32.00 avg

Complete benchmark

Run ledger.

All Guesser and benchmark-support activity.

Total tokens: 4,617,336
Wall-clock runtime: 1h 57m
Guesser calls: 562
Completed: Sep 30, 2026, 4:32:07 AM

Reliability

Output contract clean.

All 562 evaluated outputs matched the public structured-action contract.

Run details

Execution: BX-20260930-B-0003-official-M0032-001
Model ID: M-0032
Benchmark: B-0003
Base seed: 0
Git commit: 4226b3254775

Recorded cost

By role.

Cost composition.Share of full-run cost

Guesser39.6%$1.38
Primary Oracle29%$1.01
Reviewer15.8%$0.55
Judge15.3%$0.53
Validator0.2%$0.01

Excluded repair overhead$0.373 superseded attempts across 3 affected trials. Not included above.

Roles in this cost:

Guesser: Asks the questions and submits the scored guess.
Primary Oracle: Searches for evidence and proposes an answer.
Reviewer: Checks each Oracle YES or NO independently.
Judge: Decides when the Oracle and Reviewer disagree.
Validator: Checks a submitted guess against the trusted subject.

Read the role and answer-checking method →

Run configuration

Models.

Model identity, prompt versions, routing, calls, and recorded cost for the full run.

GuesserModel under test

anthropic/claude-sonnet-5.5

Asks the questions and submits the scored guess.

Calls: 562
Cost: $1.3796
Reasoning: High
Routing: Exact provider · anthropic
Resolved model: anthropic/claude-sonnet-5.5
Resolved provider: Anthropic
Prompt contract: stateful-category-guesser-v16-five-answer-category-guide
Configuration: M-0032

Resolved provider details

Anthropic562 calls$1.37961,653.86 s

Fallback calls: 0
Provider unreported: 0

Game support

Answering and checking models.

These models support the game. They are fixed across the run and are not under test.

Primary Oracle

openai/gpt-5.6-luna

Searches for evidence and proposes an answer.

Calls: 410
Cost: $1.0093
Reasoning: Medium
Routing: Exact provider · openai
Resolved model: openai/gpt-5.6-luna
Resolved provider: OpenAI
Prompt contract: live-web-oracle-v17-labelled-source-context

Resolved provider details

OpenAI410 calls$1.00933,377.24 s

Fallback calls: 0
Provider unreported: 0

Reviewer

google/gemini-3.5-flash-lite

Checks each Oracle YES or NO independently.

Calls: 394
Cost: $0.5500
Reasoning: Medium
Routing: Exact provider · google-ai-studio
Resolved provider: Google AI Studio

Resolved provider details

Google AI Studio394 calls$0.5500682.81 s

Fallback calls: 0
Provider unreported: 0

Judge

anthropic/claude-opus-5

Decides when the Oracle and Reviewer disagree.

Calls: 24
Cost: $0.5341
Reasoning: Medium
Routing: OpenRouter automatic routing
Resolved provider: Claude Platform on AWS

Resolved provider details

Claude Platform on AWS24 calls$0.5341142.77 s

Fallback calls: 0
Provider unreported: 0

Guess Validator

openai/gpt-5.6-luna

Checks a submitted guess against the trusted subject.

Calls: 43
Cost: $0.0083
Reasoning: Medium
Routing: Exact provider · openai
Resolved model: openai/gpt-5.6-luna
Resolved provider: OpenAI
Prompt contract: strict-guess-validator-v2-generic-kinds
Configuration: gpt-5.6-luna-validator

Resolved provider details

OpenAI43 calls$0.008377.91 s

Fallback calls: 0
Provider unreported: 0