Benchmark badges

Drop a live-updating “ranked #N on [benchmark]” badge on your GitHub README, HuggingFace model card, landing page, or PR description. Auto-updates whenever CodeSOTA ingests a new result — no republishing.

URL format

https://codesota.com/embed/[benchmark-id]/[model-id]

Returns an SVG badge with Content-Type: image/svg+xml, cached at the CDN for one hour. The benchmark and model IDs match the ones used across CodeSOTA — you can find them on any leaderboard URL. If the model isn't on that benchmark yet, the badge shows a polite “not on benchmark” state instead of breaking.

Color coding

The right-hand pill color reflects the model's rank on the benchmark.

  • #1 — reigning SOTA
  • Top 3
  • Top 10
  • Top 25
  • Everyone else

Live examples

Each badge below is served from the real URL — click “Copy Markdown” style snippets to drop into your README.

Claude Opus 4.6 on GPQA DiamondClaude Opus 4.6 on GPQA Diamond
Markdown
[![Claude Opus 4.6 on GPQA Diamond](https://codesota.com/embed/gpqa/claude-opus-4-6)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="Claude Opus 4.6 on GPQA Diamond" src="https://codesota.com/embed/gpqa/claude-opus-4-6"></a>
Gemini 3 Pro on GPQA DiamondGemini 3 Pro on GPQA Diamond
Markdown
[![Gemini 3 Pro on GPQA Diamond](https://codesota.com/embed/gpqa/gemini-3-pro)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="Gemini 3 Pro on GPQA Diamond" src="https://codesota.com/embed/gpqa/gemini-3-pro"></a>
GPT-5 on LiveCodeBench ProGPT-5 on LiveCodeBench Pro
Markdown
[![GPT-5 on LiveCodeBench Pro](https://codesota.com/embed/livecodebench-pro/gpt-5)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="GPT-5 on LiveCodeBench Pro" src="https://codesota.com/embed/livecodebench-pro/gpt-5"></a>
o3 on MMLUo3 on MMLU
Markdown
[![o3 on MMLU](https://codesota.com/embed/mmlu/o3)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="o3 on MMLU" src="https://codesota.com/embed/mmlu/o3"></a>
Gemini 3 Pro on Humanity's Last ExamGemini 3 Pro on Humanity's Last Exam
Markdown
[![Gemini 3 Pro on Humanity's Last Exam](https://codesota.com/embed/hle/gemini-3-pro)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="Gemini 3 Pro on Humanity's Last Exam" src="https://codesota.com/embed/hle/gemini-3-pro"></a>
Claude Opus 4.5 on Tau2-BenchClaude Opus 4.5 on Tau2-Bench
Markdown
[![Claude Opus 4.5 on Tau2-Bench](https://codesota.com/embed/tau2-bench/claude-opus-45)](https://codesota.com/browse)
HTML
<a href="https://codesota.com/browse"><img alt="Claude Opus 4.5 on Tau2-Bench" src="https://codesota.com/embed/tau2-bench/claude-opus-45"></a>

Who this is for

  • Model authors who want a trust signal on their HuggingFace card or GitHub repo
  • Papers with a release — embed in the README so the badge updates automatically as new models join the leaderboard
  • Inference providers showcasing where a hosted model ranks
  • Blog posts and launch announcements that want a compact, live rank indicator

Where the numbers come from

Every badge is backed by the same leaderboard data surfaced on /tasks and the individual benchmark pages. When a result is added, updated, or re-verified, the badge reflects it within an hour (CDN cache). For benchmarks with a trust grade, the rank is computed over all recorded results; verified results take precedence over self-reported ones.