2026-07-29 15:55 UTCIn-site rewrite2 min readUpdated: 2026-07-29 15:59 UTC

Launch HN: Tokenless (YC S26) – Automatic model switching to save money

Tokenless is a smart router that cuts AI inference costs by half by fanning out requests to multiple models and canceling the rest once one is on the right track, without sacrificing quality.

SourceHacker News AIAuthor: rohaga

Article intelligence

EngineersAdvanced

Key points

Drop-in replacement for OpenAI and Anthropic API calls, automatically routing to the cheapest model that can do the job.
Parallel request execution: sends to multiple models, then cancels all but the one that starts producing the correct answer.
Benchmarks show up to 57% cost reduction on coding tasks while matching the quality of frontier models like Opus 4.8.
Built by AI researchers from Google DeepMind, Princeton, and UC Berkeley; backed by Y Combinator.

Why it matters

This matters because drop-in replacement for OpenAI and Anthropic API calls, automatically routing to the cheapest model that can do the job.

Technical impact

May affect model selection, inference cost, product capability, and evaluation benchmarks.

This panel is AI-generated and reviewed for accuracy.

Tokenless

The router that cuts your inference bill in half.

A drop-in replacement for your API calls — always routed to the right model.

Book a demoSign up

Same quality, half the cost.

Most calls don’t need a frontier model. Tokenless fans out your request to a group of models and watches them think. Once a model is clearly on track, we select it and cancel the other models, and you only pay for what you need.

We expose an OpenAI and Anthropic compatible endpoint. Point your models at us and get started today!

Get Started

>refactor CLI options into an enum

this request$0.0000

$0.0000

sent to Fable 5$0.0000

$0.0000

−52% · $0.0101 never billed

Measured, not marketed.

Cost versus quality on public coding benchmarks — the same quality as Opus 4.8, at a fraction of the cost per task.

ModelSolved$ / taskvs Cheap.

svg]:h-3 [&>svg]:w-3" style="border-radius:0.3125rem;background:var(--color-pro-soft);color:var(--color-pro)">PRO72%$0.32-57%

GPT 5.576%$0.74—

Opus 4.872%$2.41—

svg]:h-3 [&>svg]:w-3" style="border-radius:0.3125rem;background:var(--color-max-soft);color:var(--color-max)">MAX82%$6.70—

svg]:h-3 [&>svg]:w-3" style="border-radius:0.3125rem;background:var(--color-pro-soft);color:var(--color-pro)">PROGPT 5.5Opus 4.8svg]:h-3 [&>svg]:w-3" style="border-radius:0.3125rem;background:var(--color-max-soft);color:var(--color-max)">MAX

TerminalBench 2.1

Solved72%76%72%82%

$ / task$0.32$0.74$2.41$6.70

vs Cheapest-57%———

LiveCodeBench

Solved89.0%85.7%88.9%90.4%

total $$74.78$84.27$80.01$85.33

vs Cheapest-7%———

Need maximum quality instead of maximum savings? MAX routes each task to the best available model.

See what you’d save.

YOUR MONTHLY LLM SPEND

YOUR MONTHLY BILL

$14K/mo

OFF YOUR BILL*

you pay $26K/mosaved $14K/mo

$0$40K/mo — your bill today

YOUR NEXT 12 MONTHSTREND: RAMP AI INDEX · +11.0%/MO

NEW MONTHLY BILL

$26K/mo

BLENDED SAVINGS RATE

34%

REQUESTS REROUTED

42%

CURRENT BILL

$40K/mo

Estimates use published token prices (including cache rates) and editable routing assumptions in the source. Your rate depends on your traffic. Spend history & trend: Ramp AI Index — AI spend per employee, Jun 2025–Jun 2026, matched to the cohort your spend sits in and extrapolated at its trailing 12-month growth.

Built by AI researchers from Google DeepMind, Princeton, and UC Berkeley. Backed by Y Combinator.

Cut the bill. Keep the quality.

Book a demo and we’ll run the numbers on your actual traffic — or swap two lines and see for yourself.

Book a demoSign up