已更正 · 74.01%。System Card changelog dated 2026-03-06 states that an improved cheating-detection pipeline flagged nine additional unintended solutions; the best-scoring single-agent configuration was re-verified, not rerun.
No thinking, max effort
设置
Thinking disabled; max effort.
未知项
Sampling settings are not specified in every evaluation.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
BrowseComp single-agent best configuration, 1,266 questions.
条件
Historical result recorded in the current System Card changelog before the 2026-03-06 correction. Main article's summary image still shows the rounded 74.7%. No rerun was performed for the single-agent correction.
更正记录 · 替代 74.72%。System Card changelog dated 2026-03-06 states that an improved cheating-detection pipeline flagged nine additional unintended solutions; the best-scoring single-agent configuration was re-verified, not rerun.
No thinking, max effort
设置
Thinking disabled; max effort.
未知项
Sampling settings are not specified in every evaluation.
工具
允许工具
推理预算
max
框架
未记录
协议
BrowseComp single-agent best configuration, 1,266 questions.
条件
Web search, web fetch, programmatic tool calling, context compaction triggered at 50K tokens, 10M total-token limit, max reasoning effort and no thinking.
已更正 · 82.07%。System Card changelog dated 2026-03-06 states 82.62% -> 82.07%; after removing leaked problems, five of eleven flagged questions remained correct.
Adaptive thinking, max effort
设置
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
未知项
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
BrowseComp multi-agent orchestrator and subagents.
条件
Historical value in the System Card changelog before the 2026-03-06 update; updated pipeline found 11 unintended solutions.
更正记录 · 替代 82.62%。System Card changelog dated 2026-03-06 states 82.62% -> 82.07%; after removing leaked problems, five of eleven flagged questions remained correct.
Adaptive thinking, max effort
设置
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
未知项
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
工具
允许工具
推理预算
max
框架
未记录
协议
BrowseComp multi-agent orchestrator and subagents.
条件
Orchestrator has no direct tools and can only spawn subagents; subagents use web search, web fetch and programmatic tool calling. Orchestrator uses 200K context per subagent, 50K compaction and a 1M total-token limit; max effort. Blacklist updated and affected questions rerun.
Standard configuration: adaptive thinking, max effort
设置
Adaptive thinking at max effort; default temperature and top_p; five-trial averages unless otherwise stated.
未知项
Context-window size is evaluation-dependent; standard configuration is 1M tokens.
工具
允许工具
推理预算
max
框架
未记录
协议
BrowseComp with web search, web fetch, programmatic tool calling and code execution.
条件
Adaptive thinking, max effort, 10M token limit, 200K context compaction, evaluation blocklist. Original page chart was replaced after the source identified a simpler methodology; prior numeric value was not given and is not reconstructed.
Adaptive thinking at max effort with web search, web fetch, programmatic tool calling, and code execution; 3M-token limit; context compaction is reported as triggering at 200K tokens when extending beyond the 1M context window.
未知项
Sampling parameters are not stated.
工具
允许工具
推理预算
max
框架
未记录
协议
BrowseComp Figure 8.9.2.A, test-time compute scaling at max effort with web search, web fetch, programmatic tool calling, and code execution.
条件
3M-token limit as labeled; §8.9.2 reports context compaction triggered at 200K tokens to extend beyond the 1M context window. The figure gives no exact per-point cost label.
Differences are calculated only for unique, active, exact values between direct predecessors when the unit, metric, and explicitly recorded configuration and conditions match.
· 6 score records for this release, including 6 exact values (may include historical records)
Claude Sonnet 4.6 · BrowseComp scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
Accuracy74.72%Reported 74.72% · Unit %
Published Unconfirmed
Corrected · 74.01%. System Card changelog dated 2026-03-06 states that an improved cheating-detection pipeline flagged nine additional unintended solutions; the best-scoring single-agent configuration was re-verified, not rerun.
No thinking, max effort
Settings
Thinking disabled; max effort.
Unknowns
Sampling settings are not specified in every evaluation.
Tools
Tool access not recorded
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp single-agent best configuration, 1,266 questions.
Conditions
Historical result recorded in the current System Card changelog before the 2026-03-06 correction. Main article's summary image still shows the rounded 74.7%. No rerun was performed for the single-agent correction.
Correction record · replaced 74.72%. System Card changelog dated 2026-03-06 states that an improved cheating-detection pipeline flagged nine additional unintended solutions; the best-scoring single-agent configuration was re-verified, not rerun.
No thinking, max effort
Settings
Thinking disabled; max effort.
Unknowns
Sampling settings are not specified in every evaluation.
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp single-agent best configuration, 1,266 questions.
Conditions
Web search, web fetch, programmatic tool calling, context compaction triggered at 50K tokens, 10M total-token limit, max reasoning effort and no thinking.
Corrected · 82.07%. System Card changelog dated 2026-03-06 states 82.62% -> 82.07%; after removing leaked problems, five of eleven flagged questions remained correct.
Adaptive thinking, max effort
Settings
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
Unknowns
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
Tools
Tool access not recorded
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp multi-agent orchestrator and subagents.
Conditions
Historical value in the System Card changelog before the 2026-03-06 update; updated pipeline found 11 unintended solutions.
Correction record · replaced 82.62%. System Card changelog dated 2026-03-06 states 82.62% -> 82.07%; after removing leaked problems, five of eleven flagged questions remained correct.
Adaptive thinking, max effort
Settings
Adaptive thinking enabled; max effort; default temperature and top_p unless an evaluation specifies otherwise.
Unknowns
Context window is evaluation-dependent; Table 2.1.A states it does not exceed 1M tokens.
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp multi-agent orchestrator and subagents.
Conditions
Orchestrator has no direct tools and can only spawn subagents; subagents use web search, web fetch and programmatic tool calling. Orchestrator uses 200K context per subagent, 50K compaction and a 1M total-token limit; max effort. Blacklist updated and affected questions rerun.
· 3 score records for this release, including 3 exact values (may include historical records)
Records for the direct predecessor do not provide matching conditions for a valid difference.
Claude Sonnet 5 · BrowseComp scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
Accuracy84.7%Reported 84.7% · Unit %
Published Unconfirmed
Standard configuration: adaptive thinking, max effort
Settings
Adaptive thinking at max effort; default temperature and top_p; five-trial averages unless otherwise stated.
Unknowns
Context-window size is evaluation-dependent; standard configuration is 1M tokens.
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp with web search, web fetch, programmatic tool calling and code execution.
Conditions
Adaptive thinking, max effort, 10M token limit, 200K context compaction, evaluation blocklist. Original page chart was replaced after the source identified a simpler methodology; prior numeric value was not given and is not reconstructed.
Adaptive thinking at max effort with web search, web fetch, programmatic tool calling, and code execution; 3M-token limit; context compaction is reported as triggering at 200K tokens when extending beyond the 1M context window.
Unknowns
Sampling parameters are not stated.
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
BrowseComp Figure 8.9.2.A, test-time compute scaling at max effort with web search, web fetch, programmatic tool calling, and code execution.
Conditions
3M-token limit as labeled; §8.9.2 reports context compaction triggered at 200K tokens to extend beyond the 1M context window. The figure gives no exact per-point cost label.