Benchmark Almanac

REAL

amazon/nova-act-v1.0 52.7% tops its leaderboard Claude 3.7 Sonnet 41.1% leads the models we track
Not in model cards

Agents complete 112 tasks on replicas of 11 websites. State checks grade actions, while a model judge checks retrieved answers against references.

A task succeeds only when every required outcome check passes. The reported REAL Score is the task success rate.

Made by
AGI, Inc.
Released
April 2025
Kind
Web browsing
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text, Screens → Actions, Text
Interaction
Agentic
Environment
Browser
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
In views
Computer use, Agents
Strongest model tested
o3 #84 on the capability index
Last model added
May 2025
Scores refreshed
November 19th, 2026

In the authors’ words

We introduce REAL, a benchmark and framework for multi-turn agent evaluations on deterministic simulations of real-world websites. REAL comprises high-fidelity, deterministic replicas of 11 widely-used websites across domains such as e-commerce, travel, communication, and professional networking. We also release a benchmark consisting of 112 practical tasks that mirror everyday complex user interactions requiring both accurate information retrieval and state-changing actions. All interactions occur within this fully controlled setting, eliminating safety risks and enabling robust, reproducible evaluation of agent capability and reliability. Our novel evaluation framework combines programmatic checks of website state for action-based tasks with rubric-guided LLM-based judgments for information retrieval. The framework supports both open-source and proprietary agent systems through a flexible evaluation harness that accommodates black-box commands within browser environments, allowing research labs to test agentic systems without modification. Our empirical results show that frontier language models achieve at most a 41% success rate on REAL, highlighting critical gaps in autonomous web navigation and task completion capabilities. Our framework supports easy integration of new tasks, reproducible evaluation, and scalable post-training data generation, marking a significant step forward in evaluating and advancing agent capabilities.

Abstract of REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites (arXiv:2504.11543, CC0)
17
Entries on its own leaderboard, 13 models we track

amazon/nova-act-v1.0 tops the leaderboard with 52.7% 2 models share the top score of 41.1% among models released by May 2025

Best score per entry from the benchmark's own leaderboard, read on November 19th, 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. The best model we track is Claude 3.7 Sonnet, at 41.1%. This view shows only models we track, each at its best.

The newest model with a score here came out in May 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 amazon/nova-act-v1.0 REAL v1; 112 tasks ✓ 52.7%
  2. 2 Claude 3.7 Sonnet REAL v1; 112 tasks; thinking ✓ 41.1%
  3. 3 Claude Opus 4 REAL v1; 112 tasks; thinking ✓ 41.1%
  4. 4 Claude Sonnet 4 REAL v1; 112 tasks; thinking ✓ 39.3%
  5. 5 Gemini 2.5 pro (snapshot unspecified) REAL v1; 112 tasks ✓ 38.4%
  6. 6 o3 REAL v1; 112 tasks ✓ 34.8%
  7. 7 o3-mini REAL v1; 112 tasks ✓ 25%
  8. 8 DeepSeek-V3 Mar 2025, REAL v1; 112 tasks ✓ 19.6%
  9. 9 o1 REAL v1; 112 tasks ✓ 16.1%
  10. 10 o1-mini REAL v1; 112 tasks ✓ 15.2%
  11. 11 Llama 4 Maverick REAL v1; 112 tasks ✓ 12.5%
  12. 12 Grok 3 REAL v1; 112 tasks ✓ 12.5%
  13. 13 Llama 3.3 70B Instruct REAL v1; 112 tasks ✓ 10.7%
  14. 14 Gemma 3 27B REAL v1; 112 tasks ✓ 9.80%
  15. 15 OpenAI CUA (backbone unspecified) REAL v1; 112 tasks ✓ 8.00%
  16. 16 Qwen2.5-VL-32B REAL v1; 112 tasks ✓ 2.70%
  17. 17 Llama 3.1 8b instruct REAL v1; 112 tasks ✓ 1.80%

Source: Official REAL leaderboard

#EntryScoreSettingDateSource
1 amazon/nova-act-v1.0 ✓ 52.7% REAL v1; 112 tasks Official leaderboard
2 Claude 3.7 Sonnet ✓ 41.1% REAL v1; 112 tasks; thinking Feb 2025 Official leaderboard
3 Claude Opus 4 ✓ 41.1% REAL v1; 112 tasks; thinking May 2025 Official leaderboard
4 Claude Sonnet 4 ✓ 39.3% REAL v1; 112 tasks; thinking May 2025 Official leaderboard
5 Gemini 2.5 pro (snapshot unspecified) ✓ 38.4% REAL v1; 112 tasks Official leaderboard
6 o3 ✓ 34.8% REAL v1; 112 tasks Apr 2025 Official leaderboard
7 o3-mini ✓ 25% REAL v1; 112 tasks Jan 2025 Official leaderboard
8 DeepSeek-V3 Mar 2025 ✓ 19.6% REAL v1; 112 tasks Mar 2025 Official leaderboard
9 o1 ✓ 16.1% REAL v1; 112 tasks Dec 2024 Official leaderboard
10 o1-mini ✓ 15.2% REAL v1; 112 tasks Sep 2024 Official leaderboard
11 Llama 4 Maverick ✓ 12.5% REAL v1; 112 tasks Apr 2025 Official leaderboard
12 Grok 3 ✓ 12.5% REAL v1; 112 tasks Apr 2025 Official leaderboard
13 Llama 3.3 70B Instruct ✓ 10.7% REAL v1; 112 tasks Dec 2024 Official leaderboard
14 Gemma 3 27B ✓ 9.80% REAL v1; 112 tasks Mar 2025 Official leaderboard
15 OpenAI CUA (backbone unspecified) ✓ 8.00% REAL v1; 112 tasks Official leaderboard
16 Qwen2.5-VL-32B ✓ 2.70% REAL v1; 112 tasks Mar 2025 Official leaderboard
17 Llama 3.1 8b instruct ✓ 1.80% REAL v1; 112 tasks Official leaderboard
#ModelScoreSettingDateSource
1 Claude 3.7 Sonnet ✓ 41.1% REAL v1; 112 tasks; thinking Feb 2025 Official leaderboard
2 Claude Opus 4 ✓ 41.1% REAL v1; 112 tasks; thinking May 2025 Official leaderboard
3 Claude Sonnet 4 ✓ 39.3% REAL v1; 112 tasks; thinking May 2025 Official leaderboard
4 o3 ✓ 34.8% REAL v1; 112 tasks Apr 2025 Official leaderboard
5 o3-mini ✓ 25% REAL v1; 112 tasks Jan 2025 Official leaderboard
6 DeepSeek-V3 Mar 2025 ✓ 19.6% REAL v1; 112 tasks Mar 2025 Official leaderboard
7 o1 ✓ 16.1% REAL v1; 112 tasks Dec 2024 Official leaderboard
8 o1-mini ✓ 15.2% REAL v1; 112 tasks Sep 2024 Official leaderboard
9 Llama 4 Maverick ✓ 12.5% REAL v1; 112 tasks Apr 2025 Official leaderboard
10 Grok 3 ✓ 12.5% REAL v1; 112 tasks Apr 2025 Official leaderboard
11 Llama 3.3 70B Instruct ✓ 10.7% REAL v1; 112 tasks Dec 2024 Official leaderboard
12 Gemma 3 27B ✓ 9.80% REAL v1; 112 tasks Mar 2025 Official leaderboard
13 Qwen2.5-VL-32B ✓ 2.70% REAL v1; 112 tasks Mar 2025 Official leaderboard

Number one over time

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

Anthropic
Own leaderboard Not tested since May 2025

Source: Official REAL 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%) Claude 3.7 Sonnet o3-mini o1 o1-mini 0% 20% 40% 60% 80% 100% 2025 Near the ceiling (85%) Claude 3.7 Sonnet o3-mini

Source: Official REAL 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 18 months ago, in May 2025, for Claude Opus 4. 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 (17 entries).
  • Pass Models scored 13 models we track, plus 4 other entries on its leaderboard.
  • Warning Newest score The newest came 18 months ago, in May 2025, for Claude Opus 4.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o3 (#84).
  • Pass Headroom The best is 41.1%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on realevals.ai: 14 rows for models we track and 4 other entries at the last read.
  • Pass Last read November 19th, 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