SPOT TAPE
FRNT-200K2.180+1.42%
MID-128K0.4204−0.81%
OW-70B0.0840+3.11%
OW-8B0.0191+0.44%
VIS-1M1.6400−2.19%
RSN-XL6.9000+0.93%
EMB-S0.01100.00%
14:02:07 UTC
MARKET OPEN · CONTINUOUS 24/7

Change two lines. Pay the market price for inference.

Spot is the exchange for AI tokens, and a drop-in replacement for the OpenAI API. Point your client at us, set the most you'll pay per million tokens, and stop paying list price for a commodity.

−41%
Median saving vs. list
38 ms
Routing overhead, p50
99.95%
Fill rate at your limit
SPOT MID · $ / MILLION TOKENS
LIVE
Frontier chatFRNT-200K · us-east · 1.2M tok/s
2.1800+1.42%
General chatMID-128K · eu-west · 3.4M tok/s
0.4204−0.81%
Open weights 70BOW-70B · global · 8.9M tok/s
0.0840+3.11%
Reasoning XLRSN-XL · us-west · 0.4M tok/s
6.9000+0.93%
Embeddings, smallEMB-S · global · 22M tok/s
0.01100.00%
Indicative mids, last 500 fills. Not an offer to trade. FULL BOARD →
CLEARING FOR
Sequoia-backed AI labs Fortune 500 procurement Neocloud operators Systematic funds
THE DROP-IN

Two lines change. Nothing else does.

Spot speaks the OpenAI API. Point the client at us, add a ceiling on what you'll pay per million tokens, and ship. Median integration is under ten minutes; median saving is 41%.

python typescript go curl
DIFF
client = OpenAI(- base_url="https://api.openai.com/v1",+ base_url="https://api.spot.market/v1", api_key=os.environ["SPOT_KEY"],) resp = client.chat.completions.create( model="gpt-class-mid",+ max_price=0.28, # $/M tokens. never pay above this messages=messages, stream=True, )
FILLED AT 0.2410LIST 0.4204SAVED 42.7%
max_priceFLOAT · $/M TOKENS

A hard ceiling, enforced before the request is routed. If nothing in the book clears at or under your price, you get a clean 402 instead of a surprise invoice.

regionOPTIONAL

Pin fills to a jurisdiction when residency is in your contract. Everything else routes globally for the best price.

Everything else is identical

Streaming, tool calls, structured output, vision, batch, retries, your existing SDK version. No wrapper, no proxy to run, no rewrite.

STREAM ✓TOOLS ✓JSON MODE ✓VISION ✓BATCH ✓
WHO SIGNS OFF

One switch, three different wins.

The engineer keeps their code. The founder keeps their runway. The CFO finally gets a unit cost that maps to a public benchmark.

10 min

Integration, not migration

Two lines in your client constructor. No sidecar, no proxy to operate, no change to your evals or prompt code.

38 ms

Routing overhead, p50

p95 is 71 ms. If a venue degrades mid-stream we fail over inside your price limit and you're billed once.

31

Models behind one key

Swap instruments with a string. Quality drift is benchmarked continuously and delisted automatically, so the name keeps its meaning.

−41%

Off your largest variable cost

Median saving versus provider list price. For most AI-native companies inference is the second line on the P&L after payroll.

$0

Committed spend

No annual minimum, no reserved-capacity gamble, no stranded prepaid credits. Gross margin improves the month you switch.

1 day

From signup to production

Self-serve account, wire or card, keys in minutes. Nothing to negotiate until you want credit terms.

1

Invoice across every venue

One counterparty, one contract, one monthly statement — with every fill itemized to the second and exportable to your GL.

8 bps

The only fee

No seats, no platform fee, no markup on the fill. Falls to 3 bps at volume, and we never trade against the book.

Public

Benchmark for every dollar

Your unit cost is auditable against the published Spot index — the first defensible answer to "is this price fair?"

WHAT IT'S WORTH

A worked example at $250k / month.

Assumes the median 41% clearing discount to list and the 8 bps exchange fee. Your real number depends on instrument mix and how tight you set max_price.

Current spend / month$10,000
Cleared on Spot$5,905
Saved / month$4,095
$50,000 / monthsaves $20,476
$1,000,000 / monthsaves $409,528
$2,000,000 / monthsaves $819,056
At provider list price$250,000
Cleared on Spot, fee included$147,618
SAVED PER MONTH $102,382
Annualized$1,228,584
Open an account
MECHANICS

One book. Two sides. A printed price.

Capacity is normalized to tokens per second within a model family, context window and region — so a quote from one venue is directly comparable to a quote from another. That normalization is what makes a price possible.

01

Post an order

Name the instrument, the throughput you need, the region, and your limit price. Or take the mid and be done.

02

Spot clears it

Continuous matching across 37 venues, routed to the cheapest qualifying capacity. Every fill prints to the public tape.

03

Point your SDK

One endpoint, one key, OpenAI-compatible. Capacity is live in seconds. Cash settles T+0.

BUY SIDE

Stop paying list price for a commodity.

Buyers on Spot pay the clearing price, not the rate card. Set a maximum price per million tokens and the router simply never fills above it.

Median saving versus provider list−41%
Routing overhead, p5038 ms
Fill rate at limit price99.95%
Start buying →
SELL SIDE

Idle accelerators are a wasting asset.

List throughput you aren't using by the hour — from a single node or a whole reserved cluster. Keep your floor price, keep your customers, get paid same-day.

Average utilization lift+27 pts
Time to first fill after listing4 min
SettlementT+0 fiat
List capacity →
LISTED INSTRUMENTS

Six families, thirty-one models.

Instruments are defined by capability class and context window, never by vendor. Swapping the underlying model never changes your contract.

FAMILYCONTEXTMODELSMID $/MDEPTH
Frontier chatHighest-capability reasoning and chat
200K62.18001.2M/s
General chatProduction workhorse, mid-tier quality
128K110.42043.4M/s
Open weights, large70B–405B, self-hostable equivalents
128K50.08408.9M/s
Open weights, small8B–13B, classification and routing
32K40.019118M/s
Vision, long contextDocument and video understanding
1M41.64000.6M/s
Embeddings & rerankRetrieval infrastructure, small and large
8K50.011022M/s
MARKET DATA

Know the price before you spend the money.

Every fill prints to a public tape. Stream it to set budgets, forecast unit economics, or watch your own effective rate against the index — no account needed to read it.

websocketrestpythoncsv
$ wscat -c wss://tape.spot.market/v1/fills < {"instrument":"MID-128K", "px":0.2410, "size_tok":184320, "region":"eu-west", "vs_list":-0.427, "ts":1755871327.114} < {"instrument":"OW-70B", "px":0.0838, "size_tok":220000, "region":"global", "vs_list":-0.391, "ts":1755871327.402}
SETTLEMENT, CUSTODY & CONTROLS

Built to pass a procurement review.

Spot never holds your model weights, your prompts, or your customers' data in the clear. Cash moves through a segregated account at a US chartered bank.

T+0

Same-day cash settlement

Fills net continuously; payout runs at 17:00 ET to your bank account. No credits, no expiry, no lock-in.

SEG

Segregated client funds

Balances are held apart from Spot's operating capital at a US chartered bank and reconciled daily.

SOC 2

Type II, plus ISO 27001

Annual audit, continuous monitoring, penetration tests twice a year. Reports available under NDA.

ZERO

No prompt retention

Payloads pass through encrypted and are never written to disk. Venues are contractually bound to the same terms.

SOC 2 TYPE II ISO 27001 GDPR · DPA ON REQUEST EU / US / APAC RESIDENCY SSO & SCIM 99.99% SLA
QUESTIONS

The ones we get asked most.

Talk to the team →

What am I actually buying?

Throughput on a model family — tokens per second, at a context window, in a region, for a term you choose. Not a specific vendor's endpoint.

How is quality held constant?

Every venue is benchmarked continuously against the family reference. Drift beyond tolerance delists them from the book automatically.

What happens if my venue fails mid-request?

The router retries against the next-best fill inside your price limit. You are billed once, at the original price.

Can I trade without routing traffic?

Yes. Financially-settled contracts on the Spot index are available to qualified participants, cash-settled against the daily print.

Is there a minimum?

No minimum commitment and no platform fee. You pay 8 basis points on the notional you trade, each side.

Who regulates this?

Physical delivery of compute is a commercial contract, not a security. Financially-settled products are offered under a separate, registered entity.

ON THE OTHER SIDE OF THE BOOK

Inference is a commodity — it should trade like one. Sell the capacity you aren't using, settle in fiat the same day.

Sell capacity →

Every commodity eventually gets a price. Tokens just got theirs.

Accounts open in a day. No minimum, no platform fee.