FutsuBetaAnthropic 98.5%×3.3
Claude Fable 5
claude-fable-5Mythos-class model from Anthropic, built for autonomous knowledge work and long-horizon coding.
Context 1M$26.40 / 1M
Token credits
One balance. Every model.
Buy Futsu tokens once and spend them on Claude, GPT, Gemini, DeepSeek and more — at published per-model multipliers, with no subscription required and no hidden fees. Tokens never expire; optional paid plans are platform features, never a token requirement.
Pricing
Bigger packs, cheaper tokens.
Credits are optional by design — bring your own keys and pay providers directly, with zero markup, forever. Packs exist for teams that prefer one prepaid balance: buy once, spend on any model at its published multiplier, discounts apply automatically — no codes, no negotiation, no expiry. The multiplier is the whole price: our margin lives inside that published number, there are no hidden fees on top, and BYOK stays zero-markup forever.
No payment today — we email your locked rate at launch, and you can walk away free. Early access runs on your own keys: zero markup, forever.
Starter10M$80 · $8.00 / 1M
Try real workloads without watching the meter.
≈ 900 agent runs or 10,000 chat messagesLock 10M at launchPopularBuild60M$384 · $6.40 / 1M −20%
A month of serious shipping for a small team.
≈ 5,400 agent runs or 60,000 chat messagesLock 60M at launchScale200M$1,120 · $5.60 / 1M −30%
Production pipelines with headroom to spare.
≈ 18,000 agent runs or 200,000 chat messagesLock 200M at launchEnterprise300M+Custom · from $4.80 / 1M −40%
Bulk volume for teams running Futsu in production.
Invoicing, SLA and a dedicated workspace setupTalk volumeEquivalents assume the ×1.0 baseline multiplier (Claude Sonnet 4.6) — the calculator below adjusts for your model. Credits go on sale at public launch; during early access you run free on your own keys.
5–19M tokens
$8.00
per 1M · list
20–49M tokens
$7.20
per 1M · −10%
50–149M tokens
$6.40
per 1M · −20%
150–299M tokens
$5.60
per 1M · −30%
300M+ tokens
$4.80
per 1M · −40%
Calculator
What will your usage cost?
Estimate in messages, agent runs and pages — not raw tokens.
You pay$26$26 on Claude Sonnet 4.6, $8.00 / 1M
You save vs list$0$0 List $$26.40 / mo without volume discount
≈
Workload typeAgent runs
Monthly volume300300 runs / month
ModelClaude Sonnet 4.6 ×1×1
The Starter pack (10M, $80) covers ~3 months of this workload — and tokens never expire. Or cut the cost −95%: ($1.32/mo).
Get early accessAssumes ~1,000 tokens per message, ~11,000 per typical agent run, ~300,000 per heavy coding session, ~600 per page. Model multipliers are published in the catalog — each is the model's blended list price relative to Claude Sonnet 4.6, the ×1.0 baseline. Estimates only — live metering reports the exact spend of every run.
Compare
One balance vs the alternatives.
| Futsu tokens | Raw provider keys | Chat subscriptions | |
|---|---|---|---|
| Models covered | Every catalog model, one top-up | A key and an invoice per provider | One vendor’s models |
| Runs in pipelines & agents | Native — every node meters itself | With your own glue code | Chat windows, not pipelines |
| Hard cost caps | Enforced mid-run | Alerts after the fact | Usage quotas, opaque |
| Spend visibility | Live, per node | Per-provider consoles, next day | None |
| Monthly fee | None | None | $20+ per seat |
| Unused budget | Never expires | n/a — postpaid | Resets every month |
| Volume discounts | Automatic, −10% to −40% | Negotiated, enterprise only | None |
Models catalog
Every model your tokens buy.
Pass the model id in a node's model field — Futsu routes the call to the upstream provider. The chip on each card is the billing multiplier: how much that model's tokens cost relative to the $8 / 1M base.
Anthropic 99.8%×1.7
Claude Opus 4.8
claude-opus-4-8Anthropic's most capable generally available model — text, image and file inputs with deep reasoning.
Context 1M$13.60 / 1M
OpenAI 100%×0.6
GPT-5.3 Codex
gpt-5.3-codexOpenAI's most advanced agentic coding model — frontier software-engineering performance, tool-use first.
Context 400K$4.80 / 1M
OpenAI 99.9%×0.6
GPT-5.5
gpt-5.5OpenAI's flagship general model: strong reasoning and writing across text, image and audio inputs.
Context 400K$4.80 / 1M
Anthropic 100%×1
Claude Sonnet 4.6
claude-sonnet-4-6The frontier workhorse — near-Opus quality on code and analysis. The ×1.0 baseline every multiplier is measured against.
Context 1M$8.00 / 1M
Google 99.7%×0.8
Gemini 3.5 Pro
gemini-3.5-proGoogle's top multimodal reasoner with the largest context window in the catalog.
Context 2M$6.40 / 1M
xAI 99.2%×1
Grok 4.1
grok-4.1Fast frontier model from xAI with strong math and real-time-knowledge performance.
Context 256K$8.00 / 1M
Alibaba 99.5%×0.5
Qwen 3.5 Max
qwen-3.5-maxLarge MoE generalist — competitive coding and multilingual quality below frontier pricing.
Context 256K$4.00 / 1M
DeepSeek 92.1%×0.15
DeepSeek V4 Pro
deepseek-v4-proLarge-scale Mixture-of-Experts (1.6T total, 49B active) with near-frontier reasoning at value pricing.
Context 1M$1.20 / 1M
Moonshot 97.8%×0.2
Kimi K3
kimi-k3Agentic open-weight MoE tuned for tool use and long-horizon autonomy.
Context 256K$1.60 / 1M
OpenAI 99.9%×0.1
GPT-5.3 Mini
gpt-5.3-miniDistilled frontier tier — the default for high-volume drafting and classification.
Context 400K$0.80 / 1M
Z-AI 99.8%×0.15
GLM 5.1
glm-5.1A major leap in coding capability, with particularly strong long-horizon task handling.
Context 200K$1.20 / 1M
Anthropic 99.9%×0.3
Claude Haiku 4.5
claude-haiku-4-5Fast, cheap glue — routing, extraction and short reviews between bigger nodes.
Context 200K$2.40 / 1M
Google 99.9%×0.15
Gemini 3.5 Flash
gemini-3.5-flashGoogle's high-efficiency multimodal model — near-Pro coding and reasoning at Flash-tier cost.
Context 1M$1.20 / 1M
DeepSeek 100%×0.05
DeepSeek V4 Flash
deepseek-v4-flashThe cheapest tokens in the catalog — bulk summarization, embedding prep and draft passes.
Context 256K$0.40 / 1M
Uptime is the share of routed requests that completed successfully over the last 30 days, measured across all upstream endpoints and refreshed with every deploy. The multiplier is the full billing story: tokens on a model cost the $8 / 1M base × its multiplier — volume discounts then apply on top, and there are no other fees. Each multiplier is the model's blended list price relative to Claude Sonnet 4.6, the ×1.0 baseline — our margin lives inside that published number.
How spend works
One balance. Published multipliers. Nothing else.
Every run draws from the same prepaid balance. A model's tokens cost the $8 / 1M base times its published multiplier — that chip on every catalog card is the entire billing story.
- No purchase fee, no platform surcharge, no per-seat meter
- Volume discounts to −40% apply on top, automatically
- Every run reports its exact metered cost, node by node
One metered run
agent run · claude-sonnet-4-611,400 tok
billing multiplier×1.0
futsu tokens drawn11,400
metered cost · list rate$0.09
Cost caps
No surprise bills. By design.
Prepaid tokens cap your spend by definition — and hard cost ceilings cap every run long before the balance does. When a pipeline hits its cap, it halts mid-flight, not in tomorrow's report.
- Caps stack: per run, workflow, project and workspace — strictest wins
- Live metering shows which node is eating the budget
- Resume with a higher cap, or rerun on a cheaper model
run #9c114e$3.97 / $4.00 cap
HALTED — next retry would exceed the cap
resume with a higher cap, or a cheaper model
The alternative
Rather hold your own keys?
BYOK is always available: connect your Anthropic or OpenAI keys, pay providers directly, and Futsu adds zero markup. Tokens exist for teams that want one balance instead of a key per provider.
No token markup
You pay the provider, not the middleman. Futsu never resells your usage at a margin — the platform and the meter are separate things.
Keys sealed in the vault
Ciphertext in your AES-GCM vault, scoped to you or your workspace, decrypted only at runtime. Plaintext, never at rest.
Caps still apply
Your keys plus hard ceilings: a runaway pipeline halts mid-run instead of surprising your provider invoice.
FAQ
The questions every buyer asks.
The prepaid unit every run draws from. One model token costs its published multiplier in Futsu tokens — Claude Sonnet 4.6 is the ×1.0 baseline, so Claude Opus 4.8 at ×1.7 turns 1,000 model tokens into 1,700 Futsu tokens. The base rate is $8 per 1M before volume discounts; the calculator translates that into messages, agent runs and pages so you never do token math.
No. Your balance sits in the workspace until you spend it — stock up during a sprint, spend it next quarter. There are no monthly resets and nothing is clawed back.
No. The multiplier on the Models tab is the entire billing story: base rate × multiplier, minus your volume discount. Our margin lives inside that published multiplier — no purchase fee, no platform surcharge, no per-seat meter on top.
It halts mid-run — before the spend, not in tomorrow's report. Caps stack per run, workflow, project and workspace, and the strictest one wins. You can resume with a higher cap or rerun on a cheaper model.
Yes — BYOK is always available with zero markup. Connect your Anthropic or OpenAI keys, pay providers directly, and Futsu adds nothing. Tokens exist for teams that want one balance across every model instead of a key per provider.
Bulk volumes from 300M tokens get the −40% rate, invoicing and a dedicated workspace setup. Pick the Enterprise card above or write to us after signing up — early access accounts can be upgraded in place.
Full documentation ships with public launch — until then, this page is the spec.
Top up once. Run everything.
Pick your pack, lock the discount, and spend it on whatever model fits the job — whenever. Tokens never expire.
Credits go on sale at public launch — during early access you run free on your own keys.