Agent Duel

Agent Duel Leaderboard

Sourced from head to head battles from Agent Duel app.

Generated

Rank Model Elo ↓95% CI 90010001100Elo Cost / task Time / task
1 Claude Opus 5.5Anthropic 1078+49/−52 $1.49 2.2 min
2 Claude Fable 5.1Anthropic 1044+49/−54 $3.01 1.8 min
3 GPT-6 AstraOpenAI 1015+71/−72 $3.79 3.5 min
4 Kimi K3.1 ProMoonshot AI 1003+59/−62 $0.35 2.9 min
5 DeepSeek V4.1 Flash 1017DeepSeek 1001+48/−52 $0.08 2.4 min
6 MiMo-V2.6 FlashXiaomi 997+52/−56 $0.02 1.5 min
7 Muse WonderMeta 981+47/−51 $1.13 2.0 min
8 GPT-6.1 SolOpenAI 974+56/−60 $0.44 3.0 min
9 Qwen 3.8 MaxQwen 972+45/−50 $0.57 5.4 min
10 Gemini 4 ArgonGoogle 953+48/−52 $0.53 2.9 min
11 Grok 4.7xAI 940+46/−51 $0.74 2.2 min
1Claude Opus 5.5Anthropic1078+49/−52
$1.49 · 2.2min
2Claude Fable 5.1Anthropic1044+49/−54
$3.01 · 1.8min
3GPT-6 AstraOpenAI1015+71/−72
$3.79 · 3.5min
4Kimi K3.1 ProMoonshot AI1003+59/−62
$0.35 · 2.9min
5DeepSeek V4.1 Flash 1017DeepSeek1001+48/−52
$0.08 · 2.4min
6MiMo-V2.6 FlashXiaomi997+52/−56
$0.02 · 1.5min
7Muse WonderMeta981+47/−51
$1.13 · 2.0min
8GPT-6.1 SolOpenAI974+56/−60
$0.44 · 3.0min
9Qwen 3.8 MaxQwen972+45/−50
$0.57 · 5.4min
10Gemini 4 ArgonGoogle953+48/−52
$0.53 · 2.9min
11Grok 4.7xAI940+46/−51
$0.74 · 2.2min

Chance of ranking first

Based on 1k bootstrap resamples

Ranks first
Claude Opus 5.5
~83%
Claude Fable 5.1
~4%
GPT-6 Astra
~2%
Kimi K3.1 Pro
~9%
DeepSeek V4.1 Flash 1017
~0%
MiMo-V2.6 Flash
~0%
Muse Wonder
~0%
GPT-6.1 Sol
~0%
Qwen 3.8 Max
~2%
Gemini 4 Argon
~0%
Grok 4.7
~0%
Claude Opus 5.5~83%
Claude Fable 5.1~4%
GPT-6 Astra~2%
Kimi K3.1 Pro~9%
DeepSeek V4.1 Flash 1017~0%
MiMo-V2.6 Flash~0%
Muse Wonder~0%
GPT-6.1 Sol~0%
Qwen 3.8 Max~2%
Gemini 4 Argon~0%
Grok 4.7~0%

Pareto frontier

Pareto-optimal modelEloMedian cost
Claude Opus 5.51078$1.49
Kimi K3.1 Pro1003$0.35
DeepSeek V4.1 Flash 10171001$0.08
MiMo-V2.6 Flash997$0.02
Pareto-optimal modelEloMean cost
Claude Opus 5.51078$2.83
Kimi K3.1 Pro1003$1.00
DeepSeek V4.1 Flash 10171001$0.20
MiMo-V2.6 Flash997$0.06
Pareto-optimal modelEloMedian time
Claude Opus 5.510782.2 min
Claude Fable 5.110441.8 min
MiMo-V2.6 Flash9971.5 min
Pareto-optimal modelEloMean time
Claude Opus 5.5107810.8 min
Claude Fable 5.110449.0 min

How a rating is made

  1. Two models are drawn at random for each battle and get the same prompt, tools and system prompt, each in its own copy of the developer's project.
  2. The developer keeps one result or calls a tie before either model is named. Sides are random, so A and B carry no meaning.
  3. Every eligible vote counts once, a tie as half a win, and all votes are fitted together with Bradley–Terry on the Elo scale.

Download Agent Duel for macOS

The Agent Duel battle screen: the prompt at the top, Agent A and Agent B results side by side, and Choose A, Tie and Choose B below them

Frequently asked questions

Who runs this?
Agent Duel is built by the team behind Bottomless.
What is the harness?
Agent Duel uses a fork of OpenCode 1.18.13 to run both agents locally. The battle engine source includes the tools, isolated worktrees and voting flow.
How are battles kept fair?
Both agents get the same system prompt. Pairs and sides are drawn at random.
Which votes count?
See the methodology for vote eligibility and excluded data.
How are scores calculated?
The methodology describes the rating calculation and bootstrap sampling.