Benchmark suiteLatest source data: Oct 16, 2026Checked: October 18, 2026

Agent and tool-use rankings.

A pragmatic agent-readiness view using the source-backed signals we have today: search, document work, code performance, and latency. BFCL and Terminal-Bench are called out as planned feed integrations.

Ranked models

153

Search, docs, code, or latency signal

Search Arena

42

Models with public search scores

Docs Arena

44

Models with document scores

Top proxy

Gemma 4 26b A4B

100 agent proxy

Agent formula

Workflow signals without overstating tool calling.

This page is intentionally labeled as a proxy until BFCL and Terminal-Bench are wired. It ranks search, docs, coding, and latency signals that matter for agent workflows without claiming they are a full agent benchmark.

Agent workflow proxy

Higher proxy index is better. BFCL remains a planned feed, not a silent placeholder.

Top 30 of 153

#1
Gemma 4 26b A4B

gemma-4-26b-a4b

GoogleOpen weights
100agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
0.62slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#2
Mistral Medium 3.5

mistral-medium-3.5

MistralOpen weights
100agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
0.64slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#3
Qwen3.5 122b A10B

qwen3.5-122b-a10b

AlibabaOpen weights
99agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
1.01slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#4
Claude Opus 5.5

Claude Opus 5.5 (High)

AnthropicProprietary
96agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
4.22slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#5
Claude Opus 4.6 High

claude-opus-4-6-high

AnthropicProprietary
96agent proxy
Search
Not listedSearch Arena
Docs
1,495Docs Arena rank #2
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#6
GPT-6.1 Sol Max

gpt-6.1-sol-max

OpenAIProprietary
96agent proxy
Search
1,250Search Arena rank #2
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#7
GPT-5.6 Sol xHigh

gpt-5.6-sol-xhigh

OpenAIProprietary
96agent proxy
Search
1,258Search Arena rank #1
Docs
1,489Docs Arena rank #8
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#8
Claude Opus 4.6

claude-opus-4-6

AnthropicProprietary
95agent proxy
Search
Not listedSearch Arena
Docs
1,494Docs Arena rank #4
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#9
GPT-6 Astra

GPT-6 Astra (Max)

OpenAIProprietary
94agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
1.99slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#10
Claude Opus 4.7 High

claude-opus-4-7-high

AnthropicProprietary
94agent proxy
Search
Not listedSearch Arena
Docs
1,493Docs Arena rank #5
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#11
Claude Fable 5.1 Max

claude-fable-5.1-max

AnthropicProprietary
94agent proxy
Search
Not listedSearch Arena
Docs
1,493Docs Arena rank #6
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#12
GPT-6.1 Sol

GPT-6.1 Sol (Max)

OpenAIProprietary
92agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
1.88slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#13
Claude Fable 5.1

Claude Fable 5.1 (Max)

AnthropicProprietary
90agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
1.39slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#14
Claude Fable 5.1 High

claude-fable-5.1-high

AnthropicProprietary
87agent proxy
Search
1,233Search Arena rank #4
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#15
Claude Opus 4.8 High

claude-opus-4-8-high

AnthropicProprietary
87agent proxy
Search
Not listedSearch Arena
Docs
1,486Docs Arena rank #9
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#16
Claude Sonnet 5.5

Claude Sonnet 5.5 (xHigh)

AnthropicProprietary
87agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
0.72slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#17
GPT-6 Astra Max

gpt-6-astra-max

OpenAIProprietary
86agent proxy
Search
1,246Search Arena rank #3
Docs
1,473Docs Arena rank #17
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#18
Claude Fable 5 High

claude-fable-5-high

AnthropicProprietary
85agent proxy
Search
1,229Search Arena rank #6
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#19
GPT-6 Sol Max

gpt-6-sol-max

OpenAIProprietary
85agent proxy
Search
1,228Search Arena rank #7
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#20
GPT-5.5 Search

gpt-5.5-search

OpenAIProprietary
84agent proxy
Search
1,226Search Arena rank #8
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#21
Claude Opus 4.7

claude-opus-4-7

AnthropicProprietary
83agent proxy
Search
1,210Search Arena rank #12
Docs
1,494Docs Arena rank #3
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#22
Claude Opus 5 High

claude-opus-5-high

AnthropicProprietary
83agent proxy
Search
1,214Search Arena rank #10
Docs
1,490Docs Arena rank #7
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#23
Claude Opus 4.6 Search

claude-opus-4-6-search

AnthropicProprietary
82agent proxy
Search
1,222Search Arena rank #9
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#24
GPT-5.5 High

gpt-5.5-high

OpenAIProprietary
81agent proxy
Search
Not listedSearch Arena
Docs
1,480Docs Arena rank #11
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#25
Claude Sonnet 4.6

claude-sonnet-4-6

AnthropicProprietary
79agent proxy
Search
Not listedSearch Arena
Docs
1,478Docs Arena rank #13
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#26
Claude Fable 5

Claude Fable 5 (High)

AnthropicProprietary
78agent proxy
Search
Not listedSearch Arena
Docs
1,499Docs Arena rank #1
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#27
Qwen3.8 Max

qwen3.8-max

AlibabaProprietary
78agent proxy
Search
1,229Search Arena rank #5
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#28
Kimi K3

Kimi K3 (Max)

Moonshot AIProprietary
77agent proxy
Search
Not listedSearch Arena
Docs
Not listedDocs Arena
Latency
2.97slatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#29
Claude Sonnet 5 High

claude-sonnet-5-high

AnthropicProprietary
77agent proxy
Search
Not listedSearch Arena
Docs
1,476Docs Arena rank #15
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena
#30
Claude Opus 5.5 Medium

claude-opus-5.5-medium

AnthropicProprietary
76agent proxy
Search
1,211Search Arena rank #11
Docs
Not listedDocs Arena
Latency
Not listedlatency
Tool use
BFCL pendingtool calling feed
Search ArenaDocs ArenaCode Arena

Benchmark guide

What the scores mean.

A quick reading key for agent-oriented benchmark signals, especially where tool-use benchmarks are still source-ready rather than live in the table.

Higher: agent proxyBFCL + Terminal-Bench planned
What does the agent proxy measure?

The agent proxy blends available search, document, coding, and latency signals that matter for multi-step workflows. It is useful for orientation, but it is not a full tool-use benchmark until BFCL and Terminal-Bench are wired into the data layer.

Why is BFCL not part of the score yet?

BFCL is specifically relevant to function calling and tool use, but the page only includes benchmark families once we have a reliable, auditable feed. Until then, BFCL is treated as a planned source rather than an implied score.

Why do search and document scores matter for agents?

Agents often need to retrieve information, inspect documents, and act on context before producing an answer. Search Arena and Document Arena are proxy signals for those parts of the workflow.

How should latency be interpreted here?

Latency is time to first token. Lower latency can make iterative agent loops feel faster, but it should be balanced against the model's task quality and tool reliability.