Pricing
One uncaught agent loop of 2,500 Claude Sonnet calls costs $1,200. Starter costs $99 a month. The arithmetic is below.
| Growth platform feeNov 1 to Nov 30 | $499.00 |
| Managed requests4,212,880 x $0.10 / 10k | $42.13 |
| Breakers tripped3 sessions | $0.00 |
| Fallback reroutes18,402 requests | $0.00 |
Loop spend stopped
$3,610
Return on invoice
6.7x
Every plan starts with a 14-day free production trial. No card required.
Early-stage AI SaaS under $5k/month LLM spend
$948 per year · $0.15 per 10k managed requests
Start free trialScaling SaaS with $5k to $30k/month LLM spend
$4788 per year · $0.10 per 10k managed requests
Start free trialHigh-volume AI platforms above $30k/month spend
Volume pricing
Talk to salesMetered usage
Provider tokens are never resold or marked up. You keep your keys and your provider contracts; PALM SOUQ charges for governing the traffic.
Starter fits this volume at $114 a month on monthly billing.
One caught loop on Claude Sonnet 4.5 covers 12.1 months of Starter at $99/month.
Break-even arithmetic
A coding agent retries a git merge resolution 2,500 times in 45 minutes. Each call carries 150,000 input tokens and 2,000 output tokens.
The tool thrashing heuristic stops the session after 15 identical calls. Switch the model to see how the same loop scales with list prices from the LLM Price Index.
Compare plans
Loop protection and per-tenant attribution are on every plan. Growth adds billing sync and cache accounting; Scale adds isolation and contracts.
| Feature | Starter | Growth | Scale |
|---|---|---|---|
| Gateway & providers | |||
| OpenAI, Anthropic, Google Gemini, Azure OpenAI | |||
| AWS Bedrock, DeepSeek, Groq, Mistral | |||
| Added latency at p50 | <5 ms | <5 ms | <5 ms |
| Edge points of presence | 280+ | 280+ | 280+ plus dedicated |
| Fail-open Python and TypeScript SDKs | |||
| Multi-region routing | Nearest PoP | Pinned regions | Custom topology |
| Gateway SLA | 99.9% | 99.99% | 99.999% contractual |
| Metering & attribution | |||
| Per-tenant token and cost attribution | |||
| Tenants | 50 | 1,000 | Unlimited |
| Managed requests per month | 1M | 10M | Unlimited |
| Metering retention | 7 days | 13 months | Custom |
| Prompt cache read and write tracker | |||
| Feature and user level attribution headers | |||
| Circuit breakers | |||
| Cyclical tool thrashing detection | |||
| Token velocity surge detection | |||
| Semantic echo loop detection | |||
| Max step and recursion depth limits | |||
| Per-tier and per-tenant thresholds | Per tier | ||
| Breaker detection time | <50 ms | <50 ms | <50 ms |
| Quotas & fallbacks | |||
| Soft and hard budget caps | |||
| Sliding-window rate limits per tier | |||
| Custom fallback chains | One fallback model | ||
| Provider outage rerouting | |||
| Custom routing logic | |||
| Billing & alerts | |||
| Slack and webhook alerts | |||
| Stripe, Metronome, Orb and Lago usage sync | |||
| Gross margin per tenant | |||
| OpenTelemetry export | |||
| Security & deployment | |||
| Zero Payload Retention | |||
| SOC 2 Type II report | |||
| SSO (SAML, OIDC) | |||
| HIPAA BAA | |||
| Single-tenant VPC or self-hosted gateway | |||
| Support | Email, business hours | 24x7 | 24x7 plus named FinOps architect |
“The easiest ROI calculation we have ever made. It paid for itself many times over in month one.”
Under 5 ms at p50 on every plan. Requests terminate at the nearest of 280+ edge points of presence, policy is evaluated in memory, and the call is forwarded to the provider over a pooled connection. Starter, Growth and Scale run on the same gateway; Scale adds dedicated clusters.
The Python and TypeScript SDKs fail open. If the gateway does not answer within the configured timeout, the SDK calls the provider directly with your key and backfills metering when the gateway is reachable again. Your users never see an error caused by PALM SOUQ.
Any call to a model endpoint that passes through the gateway: chat, completions, embeddings and tool calls. Requests blocked by a breaker, a rate limit or a hard cap still count, because the gateway did the work of stopping them. Health checks and dashboard API calls do not count.
Starter adds $0.15 per 10k managed requests and Growth adds $0.10 per 10k managed requests on top of the base fee, up to each plan's request ceiling. Past the ceiling we contact you about the next plan; we never stop proxying your traffic because of a PALM SOUQ plan limit.
Yes. Starter routes to the nearest PoP. Growth can pin metering and traffic to US, EU or India regions for data residency. Scale supports custom topologies, including a dedicated cluster in your own VPC.
Annual billing saves 20% on Starter and Growth. Scale contracts are priced on volume, from $1,500 a month, and include committed-use discounts above 50M requests a month.
No. You keep your own provider keys and contracts, and providers bill you directly. PALM SOUQ meters and governs that spend; it does not resell tokens or mark them up.
Yes. Upgrades apply immediately and are prorated. Downgrades take effect at the next billing date so tenant limits do not change under your users mid-month.
Fourteen days of production use on Growth features, 100k managed requests and all four loop heuristics, with no card required. At the end you pick a plan or the gateway switches to pass-through mode so nothing breaks.
Route your agent traffic through PALM SOUQ in 2 minutes. Change one base URL, pass a tenant header, and every token is attributed and governed.
14-day free production trial · 100k requests included · no card required