Benchmark Almanac

Vision2Web

Claude Fable 5 82.3% leads since Jun 2026 In use

Agents build websites from 918 prototype images across 193 tasks, from static pages to full-stack apps. A GUI agent and a VLM judge score them.

In the authors’ words

Recent advances in large language models have improved the capabilities of coding agents, yet systematic evaluation of complex, end-to-end website development remains limited. To address this gap, we introduce Vision2Web, a hierarchical benchmark for visual website development, spanning from static UI-to-code generation, interactive multi-page frontend reproduction, to long-horizon full-stack website development. The benchmark is constructed from real-world websites and comprises a total of 193 tasks across 16 categories, with 918 prototype images and 1,255 test cases. To support flexible, thorough and reliable evaluation, we propose workflow-based agent verification paradigm based on two complementary components: a GUI agent verifier and a VLM-based judge. We evaluate multiple visual language models instantiated under different coding-agent frameworks, revealing substantial performance gaps at all task levels, with state-of-the-art models still struggling on full-stack development.

Abstract of Vision2Web: A Hierarchical Benchmark for Visual Website Development with Agent Verification (arXiv:2603.26648, CC0)
Made by Z.ai (zai-org) Released Mar 2026 Web and app front ends All tasks public Last in a launch post Sep 2026 Last model added Sep 2026 Scores refreshed 2 Nov 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Open-ended Free-form text, documents, designs or media judged as a whole.
zai-org/Vision2Web 84 stars Paper Leaderboard vision2web-bench.github.io
10
Models scored on its own leaderboard
5
Launch posts list it, from 2 labs
1
System card reports it, from 1 lab

Claude Fable 5 leads with 82.3% among models released by Aug 2026

Best score per model from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in Aug 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 82.3 Claude Fable 5
  2. 80.2 Claude Opus 5
  3. 79.0 Kimi K3
  4. 77.8 GLM-5.3-Flash
  5. 76.5 Gemini 3.7 Flash
  6. 76.1 Claude Opus 4.8
  7. 73.5 GPT-5.6 Sol
  8. 67.6 DeepSeek-V4-Flash-Vision-Exp
  9. 67.5 GPT-5.6 Terra
  10. 64.5 GPT-5.6 Luna
  1. 1 Claude Fable 5 82.3%
  2. 2 Claude Opus 5 80.2%
  3. 3 Kimi K3 79%
  4. 4 GLM-5.3-Flash 77.8%
  5. 5 Gemini 3.7 Flash 76.5%
  6. 6 Claude Opus 4.8 76.1%
  7. 7 GPT-5.6 Sol 73.5%
  8. 8 DeepSeek-V4-Flash-Vision-Exp 67.6%
  9. 9 GPT-5.6 Terra 67.5%
  10. 10 GPT-5.6 Luna 64.5%
#ModelScoreSettingDateSource
1 Claude Fable 5 82.3% Claude Code Jun 2026 Official leaderboard
2 Claude Opus 5 80.2% Claude Code Jul 2026 Official leaderboard
3 Kimi K3 79% Claude Code Jul 2026 Official leaderboard
4 GLM-5.3-Flash 77.8% Claude Code Aug 2026 Official leaderboard
5 Gemini 3.7 Flash 76.5% Gemini CLI Aug 2026 Official leaderboard
6 Claude Opus 4.8 76.1% Claude Code May 2026 Official leaderboard
7 GPT-5.6 Sol 73.5% Codex Jul 2026 Official leaderboard
8 DeepSeek-V4-Flash-Vision-Exp 67.6% Claude Code Aug 2026 Official leaderboard
9 GPT-5.6 Terra 67.5% Codex Jul 2026 Official leaderboard
10 GPT-5.6 Luna 64.5% Codex Jul 2026 Official leaderboard

Labs report scores for 5 models in their cards. We plot every source in the charts below.

2 models have held number one since its release in Mar 2026

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Anthropic Qwen Z.ai
Own leaderboard Not tested since Aug 2026
Claude Fable 5 82.3%
Lab claims 2 leaders
GLM-5.3-Flash 77.8%

Scores over time

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

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5 Claude Opus 4.8 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Fable 5

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.

Qwen
Z.ai
DateLaunch postScoreAs the post writes it
18 Sep 2026 Qwen3.8-Omni-Flash 62.9 Vision2Web
26 Aug 2026 Qwen3.8 Flash 64 Visual web development Vision2Web
26 Aug 2026 GLM-5.3-Flash 77.8 Vision2Web
14 Aug 2026 Qwen 3.8 27B 62.9 Visual web development Vision2Web
3 Aug 2026 Qwen 3.8 Max 69 Vision2Web

One system card reports it

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
22 Sep 2026 Qwen3.8-Omni-Flash technical report 62.9 Vision2Web, mean of frontend, webpage and website, Claude Code, GPT-5.4 judge (Table 3) Capabilities

More in Coding

All 243 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Opus 5.5 93.9%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471