Leaderboard
Best. Smartest. Fastest.
Every text-generation model in the Newmen catalogue, ranked three ways and regenerated on each catalogue sync. Intelligence uses curated MMLU-Pro / code / math / GPQA scores where published; everything else is estimated from family + generation + popularity (clearly marked “est.”). Speed is best-of-provider p95 throughput from live traffic. Best Value is intelligence per dollar — and Atlas-1 wins by definition because it routes to whichever model is cheapest among those passing your eval gates.
Best Value
Atlas leads the field.
Highest intelligence-per-dollar. Atlas-1 takes the top slot because it routes across every model below it per call, picking the cheapest variant that has stayed green on your operation's evaluators.
#2
inclusionAI: Ling-2.6-flash
$0.01/M · Inclusion AI · est.
53000
value
#1
Atlas-1
routes across the catalogue per call
66250
value
#3
IBM: Granite 4.0 Micro
$0.02/M · IBM Granite · est.
29412
value
- 1
Atlas-1
routes across the catalogue per call
best of every model · eval-gated
66250
value
- 2
inclusionAI: Ling-2.6-flash
$0.01/M · Inclusion AI · est.
intel 53.0 · value 53000
53000
value
- 3
IBM: Granite 4.0 Micro
$0.02/M · IBM Granite · est.
intel 50.0 · value 29412
29412
value
- 4
Meta: Llama 3.1 8B Instruct
$0.02/M · Meta · est.
intel 54.5 · value 27250
27250
value
- 5
Mistral: Mistral Nemo
$0.02/M · Mistral · est.
intel 54.0 · value 27000
27000
value
- 6
OpenAI: gpt-oss-120b
$0.04/M · OpenAI
intel 81.3 · value 20841
20841
value
- 7
OpenAI: gpt-oss-20b
$0.03/M · OpenAI · est.
intel 62.0 · value 20667
20667
value
- 8
Meta: Llama 3.2 1B Instruct
$0.03/M · Meta · est.
intel 55.0 · value 20370
20370
value
- 9
LiquidAI: LFM2-24B-A2B
$0.03/M · Liquid · est.
intel 50.0 · value 16667
16667
value
- 10
Cohere: Command R7B (12-2024)
$0.04/M · Cohere · est.
intel 62.0 · value 16533
16533
value
How this updates
Regenerated on every sync.
The catalogue + per-provider stats + intelligence scores are written by pnpm models:sync. This page reads them at build time — there’s no extra API call, no caching to bust, no third-party service in the path. Refresh by running the sync and redeploying.
See the breakdown
Click any model.
Per-model pages show every provider that serves the model, with live latency / throughput / uptime / price / quant from upstream telemetry and (when published) the underlying MMLU-Pro / HumanEval / math / GPQA scores.
Browse all models →