Benchmark Almanac

WebArena-Verified

Muse Spark 1.1 69% leads since July 2026
In use

Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces.

It fixes WebArena's tasks, answers, and evaluators and drops LLM judges. A 258 task Hard subset also exists.

Made by
ServiceNow
Released
December 2025
Kind
Web browsing
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
GPT-5.4 #27 on the capability index
Last in a launch post
October 2026
Last model added
October 2026
7
Models scored in lab cards
5
Launch posts list it, from 3 labs
2
System cards report it, from 2 labs

Muse Spark 1.1 leads with 69%

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

  1. 69.0 Muse Spark 1.1
  2. 67.3 GPT-5.4
  3. 66.8 Qwen 3.8 Max
  4. 66.4 Claude Sonnet 4.6
  5. 64.8 Qwen 3.8 27B
  6. 38.4 LFM3-1.7B
  7. 26.8 LFM2.5-24B-A2B
  1. 1 Muse Spark 1.1 69%
  2. 2 GPT-5.4 67.3%
  3. 3 Qwen 3.8 Max 66.8%
  4. 4 Claude Sonnet 4.6 66.4%
  5. 5 Qwen 3.8 27B 64.8%
  6. 6 LFM3-1.7B 38.4%
  7. 7 LFM2.5-24B-A2B 26.8%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 Muse Spark 1.1 69% Jul 2026 Lab model card
2 GPT-5.4 67.3% Mar 2026 Lab model card
3 Qwen 3.8 Max 66.8% Aug 2026 Lab model card
4 Claude Sonnet 4.6 66.4% Feb 2026 Lab model card
5 Qwen 3.8 27B 64.8% Aug 2026 Lab model card
6 LFM3-1.7B 38.4% Oct 2026 Lab model card
7 LFM2.5-24B-A2B 26.8% Oct 2026 Lab model card

3 models have held number one in labs’ own numbers since its release in December 2025

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

Anthropic OpenAI Meta
Lab claims 3 leaders
Muse Spark 1.1 69%

Source: Labs’ model cards

Scores over time

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

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Muse Spark 1.1 GPT-5.4 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Muse Spark 1.1

Source: Labs’ model cards

2 labs put it in a launch post in the last six months

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

OpenAI
Qwen
Liquid AI
DateLaunch postScoreAs the post writes it
Oct 27, 2026 LFM3-1.7B 38.4 WebArena-Verified
Oct 13, 2026 LFM2.5-24B-A2B 26.8 WebArena Verified
Aug 14, 2026 Qwen 3.8 27B 64.8 Browser use WebArena-Verified (OSWorld scaffold)
Aug 3, 2026 Qwen 3.8 Max 66.8 WebArena-Verified
Mar 5, 2026 GPT-5.4 67.3 WebArena-Verified, DOM and screenshot-driven interaction

2 system cards report it, from 2 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Jul 9, 2026 Muse Spark 1.1 system card 69 WebArena-Verified pass@1, 812 tasks (Figure 44) Capabilities
Feb 17, 2026 Claude Sonnet 4.6 system card 66.4 WebArena-Verified full set, pass@1 Average@5, value printed on chart; hard subset 67.7 (Figure 2.18.2.A) Capabilities

More in Computer and browser use

All 69 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 5.5 81.8%
  • MobileWorld In use Agents complete 201 tasks across 20 mobile apps, including tasks that need user clarification or MCP tools. Backend checks verify success. GPT-5 55.3%
  • RecreationBench In use Agents explore a running reference app and rebuild it, across 250 tasks on five platforms. Programmatic tests and a VLM judge score the result. GPT-6 Astra 58.1%