Models
Every model you can call on gaents: the price as served, whether it is serving now, and how it scored on the tasks we have measured, with the sample size and a 95% interval. A blank means not measured, never zero. Compare routes per task →
Names are model families (show setting show.defaults: family shown, exact model id, provider and build not shown). Prices: GET https://api.gaents.com/v1/models at 2026-11-19T06:07:16Z. Serving and live routes: /v1/capacity at 2026-11-19T06:07:32Z. Machine-readable: /models/models.json.
| Alias | Input $/1M | Output $/1M | Serving | Live routes | extraction | placement | coding |
|---|---|---|---|---|---|---|---|
gaents/claude-haiku |
$0.1161 | $0.5802 | yes | 1 (min 2) | 91.7% [90.3%-93.0%], n=1607 (correct share of filled fields) | 0.83 [0.77-0.88], n=182 (family precision) | not measured |
gaents/claude-opus |
$4.6420 | $23.2100 | yes | 1 (min 2) | not measured | not measured | 86% [72%-93%], n=42 (runs whose check passes) go-to |
gaents/claude-sonnet |
$2.3210 | $11.6050 | yes | 1 (min 2) | 90.3% [88.8%-91.6%], n=1698 (correct share of filled fields) | 0.79 [0.73-0.84], n=182 (family precision) | not measured |
gaents/deepseek |
$0.3017 | $0.4410 | yes | 1 (min 2) | 90.5% [89.0%-91.8%], n=1651 (correct share of filled fields) go-to | 0.66 [0.58-0.72], n=178 (family precision) | not measured |
gaents/deepseek-flash |
$0.1044 | $0.2089 | yes | 1 (min 2) | not measured | not measured | not measured |
gaents/fable escalation only |
$9.2840 | $46.4200 | yes | 1 (min 2) | not measured | not measured | not measured |
gaents/glm-flash |
$0.1741 | $0.5802 | yes | 1 (min 2) | 88.8% [87.2%-90.2%], n=1745 (correct share of filled fields) | not measured | 79% [64%-88%], n=42 (runs whose check passes) |
gaents/gpt-oss |
$0.1741 | $0.6963 | yes | 1 (min 2) | 87.2% [85.6%-88.8%], n=1679 (correct share of filled fields) | not measured | not measured |
gaents/grok |
$1.4506 | $2.9013 | yes | 1 (min 2) | not measured | 0.80 [0.58-0.92], n=20 (family precision) | not measured |
gaents/qwen-27b |
$0.0696 | $2.5531 | yes | 1 (min 2) | 91.1% [89.5%-92.4%], n=1588 (correct share of filled fields) | not measured | not measured |
gaents/qwen-flash |
$0.1741 | $0.5454 | yes | 1 (min 2) | 89.9% [88.3%-91.3%], n=1612 (correct share of filled fields) | 0.77 [0.71-0.83], n=185 (family precision) go-to | not measured |
Tasks measured
- Company facts extraction (S1): 13 routes; quality columns: correct share of filled fields; recall (filled where the key has a value). Best go-to: gaents/deepseek (correct share of filled fields 90.5%, recall (filled where the key has a value) 91.6%; $0.561 cash per 1,000 correct fields): inside the interval of the top callable route gaents/claude-haiku (91.7%, interval 90.3%-93.0%; $0.6134 capacity per 1,000 correct fields) on every measured quality column. 0.91 times the incumbent's cost per correct output. The cheaper gaents/gpt-oss ($0.1469 cash per 1,000 correct fields) does not qualify: correct share of filled fields 87.2% is below gaents/claude-haiku's interval 90.3%-93.0%.
- Industry placement (family level): 10 routes; quality columns: family precision; family recall. Best go-to: gaents/qwen-flash (family precision 0.77, family recall 0.72; $1.204 cash per 1,000 correct family placements): inside the interval of the top callable route gaents/claude-haiku (0.83, interval 0.77-0.88; its cost: not recomputable at the corrected Haiku price: Pixel's race logged the CLI's total_cost_usd per call (its Haiku call log, usd_notional) and no token counts) on every measured quality column. The cheaper gaents/deepseek ($1.174 cash per 1,000 correct family placements) does not qualify: family precision 0.66 is below gaents/claude-haiku's interval 0.77-0.88; family recall 0.59 is below gaents/claude-haiku's interval 0.70-0.82.
- Coding tasks (agent with a harness): 2 routes; quality columns: runs whose check passes; mergeable as-is (blind review of passing diffs). Best go-to: gaents/claude-opus (incumbent stays): the top callable route (86%, interval 72%-93%; $0.9086 capacity per passing run). The cheaper gaents/glm-flash ($0.0703 cash per passing run) does not qualify: mergeable as-is (blind review of passing diffs) 44% is below gaents/claude-opus's interval 70%-100%.
Tasks not listed have not been measured on any route: no claim is made about them. The go-to rule is a setting (models.go_to_rule): Start from the top-scoring route you can call as one gaents alias (the incumbent). A cheaper callable route becomes the go-to only if, on EVERY measured quality column of the task, its score is inside or above the incumbent's 95% interval. If a cheaper route falls below on any column, the incumbent stays and the line says which column. Cost per correct output is ranked on cash plus capacity at list price; the two are shown in separate columns.