UABUnbiased AI BenchGlass box for model evals.
Every leaderboard, with receipts.
Home/Compare
Compare
Live · updated continuously
Compare · decision workspace

Compare finalists, not just two rows at a time.

This compare surface is built for shortlist decisions: shared evidence, obvious gaps, tie-heavy areas, and the current leader under a visible decision mode.
Selected models · 1
Shared benchmarks · 8
Current leader · Claude Sonnet 3.5
Workspace controls

Keep the decision frame explicit.

Mode and preset stay visible up top. Shortlist edits, saves, follows, and links stay available without running the page.

Current frameBest for this use caseCoding copilot · Combined public record
Decision mode
Preset

Keep the same finalists, change the weighting, and recompute the decision read without resetting the workspace.

Evidence mode
Shortlist models

Keep the compare set tight. Add one finalist at a time, then prune aggressively.

1 / 6 finalists
Workspace utilities

Save, follow, or copy this exact setup only after the shortlist looks right.

Claude Sonnet 3.5
Save or share this workspacefollow it, copy it, or reopen it later
Open evidence pageOpen workspaces
Decision read

Claude Sonnet 3.5 leads this compare set for coding copilot.

Read this page as a workspace for choosing between finalists, not as a universal crown. The current frame is Coding copilot under best for this use case.

Thin verified coverageThe current receipts are visible, but exact-match ranking is blocked by eligibility rules: historical_model
Leader under this modeClaude Sonnet 3.5
Closest counter-caseNo clear runner-up on this surface
Shared benchmark surface8 shared benchmarks · 0 tie-heavy
Evidence modeCombined public record
Why the leader is ahead
Coding · Chat / text
Where the evidence is thin
0 tie-heavy benchmarks and 32 missing leader rows keep this from reading as a settled public answer.
What to do next
Inspect decisive benchmarks first, then open the disagreement page or pairwise debates if the top line still feels too narrow.
Open evidence pageInspect benchmark disagreement
8 of 40 benchmarks
Text Arena
AR · rating
Text · Chat / text
1,28143.4%
n/a
Vision Arena
AR · rating
Vision · Vision understanding
1,14643.6%
n/a
Coding
LB · %
Code · Coding
64%83.9%
n/a
Reasoning
LB · %
Text · Reasoning / math / science
61.7%77.4%
n/a
Language
LB · %
Text · Chat / text
55.5%83.9%
n/a
Instruction following
LB · %
Text · Chat / text
71.1%38.7%
n/a
Coding generation
LB · %
Code · Coding
59%71%
n/a
Coding completion
LB · %
Code · Coding
69%90.3%
n/a
Decisive benchmarks

What is doing the visible work

Text Arena
AR · spread n/a · rating
Vision Arena
AR · spread n/a · rating
Coding
LB · spread n/a · %
Decision pressure

What changes the winner

Claude Sonnet 3.5
32 visible benchmark gaps still leave room for the result to move.
Coverage gaps

Where evidence is missing

Claude Sonnet 3.5
Missing visible evidence on this compare surface
32
Preset interpretation

How this weighting reads the field

The current decision mode is grounded in the Coding copilot preset. This keeps the compare page connected to a visible use case instead of an unspoken “overall winner” claim.

score = 1.45 × coding + 1.05 × reasoning math science + 0.80 × long contextcoverage floor = 55% · recency window = 120 days
Share or publish this compare resultlinks, copy, and advanced framings
Share this artifact

Publish the claim after the evidence, not before it.

Keep the receipts page and card handy, then use the advanced framings only when you actually need them. Share stays available without becoming the page.

Compare artifactClaude Sonnet 3.5 leads this compare set for coding copilot.

0 shared benchmarks are still tie-heavy, so the win stays conditional. This compare uses the combined public record, with hybrid receipts labeled separately.

Evidence page/artifacts/compare?models=claude-sonnet-3-5&mode=best-for-this-use-case&preset=coding-copilot
Card image/api/og?kind=compare&models=claude-sonnet-3-5&mode=best-for-this-use-case&preset=coding-copilot
Public links

Open or copy the stable surfaces

The receipts page is the canonical evidence surface. The card image is the compact preview for embeds, screenshots, and social cards.

Open evidence pageOpen card preview
Copy-ready text

Use the exact public framing

Each copy action keeps the claim attached to receipts instead of forcing you into a blank composer.

Advanced framings and X composerNeutral, contrarian, open-model, and skeptical variants
Compare artifact

Pick the voice before you post

Use the framing variants only when you need them. The artifact page and the public copy actions above should handle most cases.

Neutral analystLead with the claim, then attach the reason and caveat.Claude Sonnet 3.5 leads this compare set for coding copilot.
ContrarianPush against the easy read and keep the counter-case live.Contrarian take: Claude Sonnet 3.5 leads this compare set for coding copilot.
Open-model angleBias the framing toward the open-weight or transparent-evidence angle.Open-model angle: Compare artifact · Claude Sonnet 3.5
Don't trust the headlineLead with the caveat before you let the claim travel.Don't trust the headline: Compare artifact · Claude Sonnet 3.5
X composer

Compose a post that keeps the caveat attached

The post shell always exposes the headline, why, caveat, receipts link, and an optional reply-bait angle.

HeadlineClaude Sonnet 3.5 leads this compare set for coding copilot.
WhyAR · spread n/a · rating
Caveat0 shared benchmarks are still tie-heavy, so the win stays conditional. This compare uses the combined public record, with hybrid receipts labeled separately.
Receipts link/artifacts/compare?models=claude-sonnet-3-5&mode=best-for-this-use-case&preset=coding-copilot
PreviewOver 280
Claude Sonnet 3.5 leads this compare set for coding copilot.
AR · spread n/a · rating
Caveat: 0 shared benchmarks are still tie-heavy, so the win stays conditional. This compare uses the combined public record, with hybrid receipts labeled separately.
Receipts: /artifacts/compare?models=claude-sonnet-3-5&mode=best-for-this-use-case&preset=coding-copilot
Open in XOpen card preview
Reading guide

How to read this workspace

Who wins most oftenBenchmarks with one clear percentile leader.Missing evidenceBenchmarks where a finalist is absent from the visible surface.Decision readA shortlist claim tied to an explicit preset and mode.