Curated launch set

Six models, hosted well.

We host fewer models, better. Curated for license, capability, and a documented buyer profile. Every model lists best-for and not-for explicitly — picking the wrong one for your workload is what catalogs full of options reward.

// pick by job
filter
llama-3.1-70b-instruct

Llama-3.1-70B-Instruct

The workhorse. Pick this if you don't have a reason not to.
70.6B · dense
128k context
$0.60 / $0.80 per 1M
us · eu · in
best forGeneral chat, RAG, classification, moderate tool-use.
not forFrontier-class reasoning — use 405B. High volume — use FP8.
Llama 3.1 Communityfunction calling
llama-3.1-70b-instruct-fp8

Llama-3.1-70B-Instruct · FP8

40% cheaper, ~2× throughput. Pick this once you've A/B-tested quality.
70.6B · FP8
128k context
$0.35 / $0.45 per 1M
us · eu · in
best forHigh-volume summarization, RAG, classification.
not forQuality-critical work you haven't validated on FP8 yet.
Llama 3.1 Communitycost-optimised
llama-3.1-405b-instruct

Llama-3.1-405B-Instruct

The frontier flagship. Head-to-head with GPT-4o and Claude 3.5 Sonnet.
405B · dense
128k context
$3.00 / $5.00 per 1M
us · eu · in
best forComplex reasoning, long-context analysis, high-quality code.
not forCheap, high-volume traffic — use 70B or DeepSeek-V3.
Llama 3.1 Communityfrontier
hermes-3-405b

Hermes-3-405B

Agent-tuned and steerable. Read the safety section before you ship.
405B · dense
128k context
$3.00 / $5.00 per 1M
us only · beta
best forAgent workloads, security research, steerable assistants.
not forConsumer chat without your own safety layer; EU / India residency at launch.
betaLlama 3.1 Community

Looking for a different model?

We host fewer models, better. The Phase-2 candidate list and a customer-vote form for what to add next live in the dashboard once you're signed in.

Vote on the next addition →