coffee
one coffee
$5
- ›your handle in the sponsors roll
- ›keeps free-tier runs humming
fund the bench
← leaderboardFree models run on free tiers. Paid models need gas money. Pitch in and we run the evals the community wants to see.
season goal
$0 / $100
0% funded
Buy Me a Coffee →GitHub Sponsors (coming soon)
support tiers
sprint
eval sprint
$25
- ›fund one full 12-challenge run of a paid model
- ›name the model in the request queue
- ›sponsors roll + shoutout in the next results drop
season
season pass
$100
- ›fund a month of new-model drops
- ›vote on the next signature challenge
- ›name on the dashboard footer for 30 days
model wishlist — what your $ buys
| model | tier | why | est. | status |
|---|---|---|---|---|
| Claude Opus 4.8 | paid | top-shelf coding model — expensive per full 12-challenge suite | $8 | unfunded |
| Claude Sonnet 5 | paid | current frontier mid-tier; needs a fresh multi-run pass | $4 | unfunded |
| GPT-4o | paid | OpenAI flagship — community keeps asking for a head-to-head | $5 | unfunded |
| Gemini 2.5 Pro | paid | long-context coding muscle, paid quota | $3.50 | unfunded |
| Llama 3.3 70B (Groq) | free | free via Groq — just needs someone to press go | free | scored |
| Gemini 2.5 Flash | free | free AI Studio quota — replaced retired 2.0 Flash | free | scored |
| GPT-5.6 | paid | OpenAI's latest flagship — the head-to-head everyone asks for | $6 | unfunded |
| Claude Fable 5 | paid | Anthropic's Mythos-class tier above Opus — the ceiling check | $10 | unfunded |
| Claude Opus 4.6 | paid | Opus lineage baseline — anchors the 4.x curve | $7 | unfunded |
| Claude Opus 4.7 | paid | one step up the Opus 4.x curve from 4.6 | $7.50 | unfunded |
| GLM 2.5 | cheap | Zhipu's open-weight coder — cheap API, big claims | $1.50 | unfunded |
| Kimi K3 | cheap | Moonshot's agentic coder — cheap API, strong on long tasks | $2 | unfunded |
already on the board: GPT-OSS 120B · Claude Sonnet 4.6 · Claude Haiku 4.5 · Llama 3.3 70B · Gemini 2.5 Flash
how the money works
- ✓Free-tier models (Groq Llama, Gemini Flash) always run first — no card required.
- ✓Crowdfunded dollars only buy API tokens for published evals, never ads or merch.
- ✓Every funded run ships raw responses + rubric scores in results.json. Show the work.
- ✓You can still fork and run paid models yourself — crowdfunding is optional fuel.
run it yourself for $0
# free subjects + free judge — no paid APIs required export GROQ_API_KEY=... export GEMINI_API_KEY=... python bench.py --tier free --judge gemini-2.0-flash --runs 3 --output results.json cd web && node scripts/sync-results.mjs && npm run dev