Lab

The research lab behind the model, with its mark. Sortable like every other column.

AA index

Artificial Analysis Intelligence Index, independent benchmark score, higher is smarter (top of their leaderboard ≈ 53). We use each model's strongest reasoning variant; the bar is scaled to 53. The thin sparkline under the bar is the value trend (index per blended $/1M at a 3:1 mix): it moves when the score changes or the price does, so a flat index with a falling sparkline means silent value decay. Hover it for first-to-last numbers. * = closest AA variant where names differ slightly. Colors: ≥40 green, 30–39 light green, 22–29 amber, below grey.

AA pos

Where that variant sits on AA's ~300-model leaderboard (e.g. #17). A relative quality rank across the whole market, not just Go.

Detail ▸

Click any row marked ▸ to expand its AA benchmark detail: Coding index, inference speed (tokens/sec), time to first token, and the model creator, with a link to its full Artificial Analysis page. Models with no AA leaderboard entry have no ▸ and nothing to expand.

Coding-verified

coding-verified badge: Artificial Analysis Coding Agent Index score, Jul 2026 snapshot. Scores are per agent+harness variant (see tooltip); AA has since updated the benchmark suite, so treat as a snapshot not a live ranking.

Go $/mo

Your monthly usage allowance in dollars for that model on the $10 Go sub, not a charge. Enforced as 5-hour (20%), weekly (50%), monthly (100%) caps. The small index-times-allowance figure under it is total smart output the plan buys.

Req / 5h

OpenCode's estimate of typical agentic requests per 5-hour window before you hit the cap. High-cached-prompt workloads go much further.

Vol

System-wide usage: token volume across all of OpenCode for that model, from the OpenCode Data top-18 ranking (e.g. 36T), with rank and week-over-week change. Refreshed from the live page each watch tick; only ranked models show a value. The sparkline traces the last 14 snapshots, with ▲▼ showing rank movement since the previous run. An amber ● on any number means it changed recently, hover for the old value and date.

In / Out $/1M

Token burn rate per million tokens at base rates. Peak windows (01:00–04:00 and 06:00–10:00 UTC, Mon–Fri; weekends off-peak) are tracked by the status bubble up top.

Sorts (default: Best value): AA rank (smartest first) · Best value (index per blended burn $/1M at a 3:1 mix, base prices) · Smarts × allow (index times $ allowance: total smart output the plan buys) · Requests/5h (biggest allowance first). Every column header sorts too, hover any header for what it measures, click again to reverse; blanks always sink to the bottom. The + on each row adds it to a side-by-side compare panel (up to 3 models, works across both tabs). Training Model = the provider trains on your prompts, today that's Meta's Contributor models (heavily discounted pricing in exchange, limited regions); any future model with the same deal gets the same tag. China-only = served from China-based infrastructure (DeepSeek); every other Chinese lab routes through international ZDR infrastructure.

Lab

The research lab behind the model, with its mark. Sortable like every other column.

AA index / pos

Same Artificial Analysis Intelligence Index and leaderboard position as the Go tab, so you can compare quality across both plans. The thin sparkline under the index is the value trend (index per blended $/1M at a 3:1 mix, hover for numbers). Click any row marked ▸ for benchmark detail (Coding index, speed, latency, creator). No ▸ means no AA leaderboard entry for that model.

Vol

Same system-wide token volume from the OpenCode Data top-18 as the Go tab. unlisted = used on OpenCode but absent from Zen's pricing page (no public per-token price), shown for volume only.

In / Out $/1M

Zen is pay-as-you-go: real charges per million tokens, no allowances. Base tier shown; tiered models cost more past a context threshold (e.g. GPT ≤272K vs above, Claude >200K).

Tags

Free = $0 for a limited feedback period (some train on data). Training Model = provider trains on your prompts (same meaning as the Go tab). † deprecated = OpenCode has dated it for removal. −50% = launch discount with expiry. China-only = served from China-based infrastructure (DeepSeek only). Unannounced eval models sit in the black STEALTH table under the main one.

Coding-verified

coding-verified badge: Artificial Analysis Coding Agent Index score, Jul 2026 snapshot. Scores are per agent+harness variant (see tooltip); AA has since updated the benchmark suite, so treat as a snapshot not a live ranking.

Sorts (default: Best value): AA rank · Best value (index per blended $/1M at a 3:1 input/output mix, AA's convention, free scored models float to the top) · cheapest input · cheapest output. Every column header sorts too, click again to reverse. The + on each row adds it to a side-by-side compare panel (up to 3 models, works across both tabs). Zen billing: add $20 balance, auto-reloads $20 below $5 (adjustable), monthly workspace caps available.