Benchmark Almanac

AgentRewardBench

o4-mini 82% leads since April 2025
Not in model cards

Models judge whether web-agent trajectories completed their tasks. Their judgments are compared with expert labels across 1,302 trajectories from five benchmarks, using precision, recall, and F1.

The model under evaluation is the judge, not the agent that produced the trajectory. Trajectories also carry annotations for side effects and repetition.

Released
April 2025
Kind
Web browsing
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Screens → Text
Interaction
Single turn
Items from
Generated
Refresh
One fixed set
Run by
Anyone
Strongest model tested
o4-mini #95 on the capability index
Last model added
April 2025
Scores refreshed
November 18th, 2026

In the authors’ words

Web agents enable users to perform tasks on web browsers through natural language interaction. Evaluating web agents trajectories is an important problem, since it helps us determine whether the agent successfully completed the tasks. Rule-based methods are widely used for this purpose, but they are challenging to extend to new tasks and may not always recognize successful trajectories. We may achieve higher accuracy through human evaluation, but the process would be substantially slower and more expensive. Automatic evaluations with LLMs may avoid the challenges of designing new rules and manually annotating trajectories, enabling faster and cost-effective evaluation. However, it is unclear how effective they are at evaluating web agents. To this end, we propose AgentRewardBench, the first benchmark to assess the effectiveness of LLM judges for evaluating web agents. AgentRewardBench contains 1302 trajectories across 5 benchmarks and 4 LLMs. Each trajectory in AgentRewardBench is reviewed by an expert, who answers questions pertaining to the success, side effects, and repetitiveness of the agent. Using our benchmark, we evaluate 12 LLM judges and find that no single LLM excels across all benchmarks. We also find that the rule-based evaluation used by common benchmarks tends to underreport the success rate of web agents, highlighting a key weakness of rule-based evaluation and the need to develop more flexible automatic evaluations. We release the benchmark at: https://agent-reward-bench.github.io

Abstract of AgentRewardBench: Evaluating Automatic Evaluations of Web Agent Trajectories (arXiv:2504.08942, CC0)
9
Entries on its own leaderboard, 4 models we track

o4-mini leads with 82% among models released by April 2025 o4-mini leads with 82% among models released by April 2025

Best score per entry from the benchmark's own leaderboard, read on November 18th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 o4-mini 82%
  2. 2 WebJudge-7B 75.7%
  3. 3 World-State-Model-7B 71.2%
  4. 4 GPT-4o Mar 2025, A 69.8%
  5. 5 Claude 3.7 S. S 69.4%
  6. 6 Llama 3.3 A 67.7%
  7. 7 Qwen2.5-VL S 64.5%
  8. 8 GPT-4o mini S 64.5%
  9. 9 Llama 3.3 70B 52.5%

Source: Official AgentRewardBench leaderboard

Source: Official AgentRewardBench leaderboard

#EntryScoreSettingDateSource
1 o4-mini 82% WebJudge Apr 2025 Official leaderboard
2 WebJudge-7B 75.7% Official leaderboard
3 World-State-Model-7B 71.2% Official leaderboard
4 GPT-4o Mar 2025 69.8% A Mar 2025 Official leaderboard
5 Claude 3.7 S. 69.4% S Official leaderboard
6 Llama 3.3 67.7% A Official leaderboard
7 Qwen2.5-VL 64.5% S Official leaderboard
8 GPT-4o mini 64.5% S Jul 2024 Official leaderboard
9 Llama 3.3 70B 52.5% NNetNav Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 o4-mini 82% WebJudge Apr 2025 Official leaderboard
2 GPT-4o Mar 2025 69.8% A Mar 2025 Official leaderboard
3 GPT-4o mini 64.5% S Jul 2024 Official leaderboard
4 Llama 3.3 70B 52.5% NNetNav Dec 2024 Official leaderboard

Rankings by part

We read 5 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as AssistantBench, VisualWebArena and WebArena. o4-mini leads 4 of them, and GPT-4o leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

AssistantBench

4 models we track have a score here.

  1. 1 o4-mini 100%
  2. 2 GPT-4o Mar 2025 83.3%
  3. 3 GPT-4o mini 80%
  4. 4 Llama 3.3 70B 20.8%

2 models have held number one on the official leaderboard since its release in April 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard Not tested since Apr 2025
o4-mini 82%

Source: Official AgentRewardBench leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o4-mini GPT-4o GPT-4o mini 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) o4-mini GPT-4o

Source: Official AgentRewardBench leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Computer and browser use

All 119 in Computer and browser use
  • BrowseComp In use Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness. Claude Fable 5 93.3%
  • ScreenSpot-Pro In use GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element. GPT-6 Astra 92.7%
  • AndroidWorld In use Agents complete 116 tasks across 20 real Android apps in an emulator. Each task checks the device's system state to judge success. GPT-6 Astra 100%
  • OSWorld 2.1 In use Agents complete 108 long-horizon workflows across 31 self-hosted websites and desktop apps, averaging 318 tool calls. Scored by binary completion and checkpoint partial credit. Claude Opus 5 44.3%
  • WebArena-Verified In use Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces. Muse Spark 1.1 69%
  • OSWorld 2.1 (offline subset) In use Agents complete the 82 OSWorld 2.1 tasks on the official offline list, with no internet access. Scored by checkpoint partial credit and strict pass rate. Claude Fable 5.2 89.3%

Data audit

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 19 months ago, in April 2025, for o4-mini. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (9 entries).
  • Pass Models scored 4 models we track, plus 5 other entries on its leaderboard.
  • Warning Newest score The newest came 19 months ago, in April 2025, for o4-mini.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o4-mini (#95).
  • Pass Headroom The best is 82%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 8 rows for models we track and 5 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check