---
type: "post"
id: "vals-reports-the-clearest-improvement-over-fable-5-fd58c3"
slug: "vals-reports-the-clearest-improvement-over-fable-5-fd58c3"
title: "Vals reports the clearest improvement over Fable 5.1 in coding."
description: "Vals AI’s Vibe Code Bench v1.1 ranks Claude Fable 5.2 first for building web apps from scratch, with a 93.77% score in OpenHands."
retrieval_nugget: "Vals AI’s Vibe Code Bench v1.1 ranks Claude Fable 5.2 first for building web apps from scratch, with a 93.77% score in OpenHands. Its lead over Fable 5.1 is modest, while the published cost and runtime figures make the result useful for teams comparing quality against execution expense."
published_at: "2026-11-10"
updated_at: "2026-11-10"
record_date: "2026-11-10"
date_kind: "published_at"
topics: ["claude","gemini"]
entities: ["Vals AI","Claude Fable 5.2","Claude Fable 5.1","Claude Sonnet 5.5","Gemini 4 Argon","OpenHands"]
related_patterns: ["agent-economics-moves-to-completed-work"]
novelty_label: "new_fact"
what_changed: "Vals’ Vibe Code Bench v1.1 shows Claude Fable 5.2 with the clearest coding improvement over Fable 5.1."
prior_refs: []
source_url: "https://www.vals.ai/benchmarks/vibe-code"
source_urls: ["https://www.vals.ai/benchmarks/vibe-code"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:vals-reports-the-clearest-improvement-over-fable-5-fd58c3"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/vals-reports-the-clearest-improvement-over-fable-5-fd58c3/","markdown":"https://newruntime.com/posts/vals-reports-the-clearest-improvement-over-fable-5-fd58c3.md","json":"https://newruntime.com/posts/vals-reports-the-clearest-improvement-over-fable-5-fd58c3.json"}
markdown_url: "https://newruntime.com/posts/vals-reports-the-clearest-improvement-over-fable-5-fd58c3.md"
---

# Vals reports the clearest improvement over Fable 5.1 in coding.

## Retrieval answer

Vals AI’s Vibe Code Bench v1.1 ranks Claude Fable 5.2 first for building web apps from scratch, with a 93.77% score in OpenHands. Its lead over Fable 5.1 is modest, while the published cost and runtime figures make the result useful for teams comparing quality against execution expense.

## Summary
Vals AI’s Vibe Code Bench v1.1 ranks Claude Fable 5.2 first for building web apps from scratch, with a 93.77% score in OpenHands. Its lead over Fable 5.1 is modest, while the published cost and runtime figures make the result useful for teams comparing quality against execution expense.

## What happened
Vals AI’s Vibe Code Bench v1.1 leaderboard, dated Nov. 09, 2026, evaluates models on building web apps from scratch. In the listed overall results, Claude Fable 5.2 ranks first with a 93.77% score, a cost of $22.13 per test, 34.71 million tokens per test and a duration of 40 minutes 37 seconds. The harness listed for the run is OpenHands. The detailed table gives the accuracy as 93.77% ±1.15.

Claude Sonnet 5.5 is second at 92.39% ±1.26, with a listed cost of $19.73 per test and duration of 1 hour 1 minute. Gemini 4 Argon ranks third at 91.91% ±1.90, costing $8.11 and taking 37 minutes 28 seconds. Claude Haiku 5.5 is fourth at 90.44% ±1.73, with a $6.07 cost and duration of 34 minutes 36 seconds. The table also places Claude Fable 5.1 seventh at 90.26% ±1.57, with a $33.37 per-test cost and a duration of 57 minutes 40 seconds.

The comparison therefore shows Fable 5.2 ahead of Fable 5.1 by 3.51 percentage points in this benchmark, while also listing lower cost and shorter duration for the newer model in these runs. Other models produce a different balance: Gemini 4 Argon is below Fable 5.2 on the score but has a much lower listed test cost, while Haiku 5.5 also combines a lower score with lower cost and duration. The leaderboard spans 127 systems and reports harness, score, cost, token use and duration; most rows use OpenHands.

Vals presents the benchmark as a measure of web-app construction from scratch. Its leaderboard includes multiple labs and models, and the same page also displays controls for filtering models, harnesses and tasks. The reported figures describe those benchmark runs; they do not establish how a model will perform on a specific company codebase or with a different toolchain.

## What the archive adds
New Runtime’s earlier coverage on Oct. 18, 2026 examined Vals AI’s SWE-bench Multimodal leaderboard and argued that benchmark leadership is conditional on the environment: [Claude Opus 5.5 tops Vals AI’s SWE-bench Multimodal at 70.21%, 9 points ahead of Claude Fable 5.1](https://newruntime.com/posts/claude-opus-5-5-tops-vals-ai-s-3c4a0e/). This is a new result on a different Vals benchmark, not a restatement: it adds a web-app-building comparison and fresh cost and duration figures for Fable 5.2.

## Trend signal
This supports the idea that benchmark leadership is useful only when read alongside the conditions and costs that produced it. Vals reports accuracy, harness, per-test cost, token use and duration, making the ranking more actionable than a score alone. The signal is consistent with New Runtime’s earlier analysis of benchmark context, not evidence that one model is universally best: [agent economics moves to completed work](https://newruntime.com/patterns/agent-economics-moves-to-completed-work/).

## Why now
The Vibe Code Bench v1.1 results were dated Nov. 09, 2026, the same date shown in New Runtime’s earlier coverage of Claude Fable 5.2 and Claude Mythos 5.2. The new leaderboard makes a task-specific comparison available alongside the model’s launch, but does not establish performance beyond its own test setup.

## Why it matters
For a team choosing a coding-agent model, a higher benchmark score is only one input. On this test, Fable 5.2 improves on Fable 5.1 while also showing a lower listed per-test cost and shorter runtime, but alternatives such as Gemini 4 Argon and Haiku 5.5 have substantially lower listed costs. That gives teams a concrete starting point for deciding what to test rather than a reason to switch blindly.

The practical lesson is to translate leaderboard claims into local evaluation. Run representative tasks through the harness and workflow you intend to use, then compare successful completion, review burden, runtime and spend. The Vals result concerns web apps built from scratch; it cannot by itself answer whether a model is more effective on an existing repository, a particular language or a team’s own acceptance criteria.

## Limits and uncertainty
Vals reports benchmark scores and resource figures, including uncertainty ranges for many accuracy results. The leaderboard does not by itself show whether those results transfer to production repositories, different prompts, other harnesses or different evaluation criteria. A benchmark score is not a measure of maintenance quality, security, or the human review needed after generation. The published figures should therefore be treated as comparative results within Vibe Code Bench v1.1, not as a complete measure of coding-agent value.

## Skill to build now
Build a small, repeatable model-evaluation set from three to five web-app tasks your team actually encounters. Fix the prompt, acceptance checks and harness, then run candidate models on the same tasks. Record pass rate, human correction time, elapsed time and API cost; this lets you test whether a leaderboard advantage survives contact with your own work.

## Reader takeaway
Fable 5.2 leads Vals AI’s web-app benchmark, but the useful comparison is not rank alone: its measured score, cost and runtime should be weighed against cheaper alternatives and checked on your own tasks.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- topic hub: [Gemini](https://newruntime.com/topics/gemini.md): Explore the gemini topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)