Looking for an alternative to Vercel AI Gateway?
Vercel AI Gateway sells tokens at provider list price from prepaid credits, with no markup, and leaves the choice of model to you. Standard Compute optimizes for cost instead: its router sends each request to the cheapest model that can do it, on providers we vet for price, uptime and speed with automatic fallback, and the compute is included in one plan per key.
TL;DR
Stay on Vercel if you want list-price tokens around an app already on Vercel. Switch if you want the cost routing done for you and each developer's month capped at a fixed price.
What Vercel AI Gateway offers
"No markup and no platform fee on tokens": pay-as-you-go credits at provider list price, plus any card processing fees; purchased credits expire after a year. Every Vercel team gets a monthly free credit on a subset of models with lower rate limits, which stops once you buy credits. BYOK has no fee. Paid add-ons include team-wide zero data retention and provider allowlists ($0.10 per 1,000 requests each). (Verified 2026-10-13.)
- Tokens at list price, with no fee on your own provider keys either
- Budgets for a team, project, API key or team member
- One-command Claude Code setup (npx vercel ai-gateway setup --agent claude-code), plus Claude Pro and Max subscription passthrough
- A natural fit for apps already on Vercel and the AI SDK; invoicing and volume discounts for enterprise
Why people look for an alternative
- List price with no markup still means a per-token bill: a heavy agent month costs list price for every token
- The free credit comes with lower rate limits (429s)
- Team-wide zero data retention and provider allowlists are paid add-ons on Pro and Enterprise
- No automatic model choice: by default it picks a provider by recent uptime and latency, but which model handles a request is yours to set
- Budgets are "a soft cap, not a hard limit", and spend on your own provider keys isn't counted by default
22 Vercel AI Gateway alternatives, grouped by what you need
List-price tokens leave one lever on the bill: which model handles each request. The strongest Vercel AI Gateway alternative for cost is a gateway that makes that choice for you, with the compute in one plan; Standard Compute, our product, is one, and its entry says when it is the wrong pick. The other groups cover spend limits that refuse requests, gateways with their own accounts, EU hosting and enterprise needs.
| Option | Best for | How you pay | Model choice and fallback | SSO and certifications | EU option |
|---|---|---|---|---|---|
| Standard Compute | Cost routing | Monthly plan from $19 per key, compute included; requests stop at the budget | Router picks the cheapest capable model, or pin a model ID; provider fallback | SSO (SAML) on Enterprise or as an add-on; no SOC 2 yet; roles and an audit log | Europe or US rules per company, team or key; requests pass through our gateway |
| Kilo Gateway | Cost routing | No token markup, 5% payment fee; or Kilo Pass | kilo-auto/efficient, by difficulty | Not published | Enterprise only |
| Cloudflare AI Gateway | Cost routing | Free layer; BYOK or credits plus 5% on purchases | Auto Router weighs cost and quality; Dynamic Routing | Dashboard SSO on all plans | Not covered by its Data Localization Suite |
| CodeRouter | Cost routing | Monthly plan counted in tokens; overage up to a cap | Picks by task phase; pinning rewritten to auto on plans | Not published | Enterprise only |
| Requesty | Spend limits | Provider price plus 5% on top-ups; BYOK | Fallback and load-balancing policies you write | SSO and audit logs on Enterprise | EU endpoint on every plan |
| TrustedRouter | Spend limits | Provider price plus 5.5% | Fixed alias lists (cheap, zdr, eu); provider fallback | SOC 2 Type II window still open | EU gateway; upstreams may process outside the EU |
| LiteLLM | Spend limits | Free; Enterprise license; you pay providers | Auto Router add-on by difficulty tier; fallbacks you configure | SSO free to 5 users; SOC 2 Type II | Wherever you host it |
| Agentgateway | Spend limits | Free (Apache 2.0); Solo.io sells support | Weighted, failover and conditional virtual models | Admin RBAC and audit logs not found in its docs | Wherever you host it |
| OpenRouter | No Vercel account | Provider price plus 5.5% on credits; BYOK free to $25,000 a month | You, or its Auto and Pareto routers; provider fallback | SSO/SAML on Enterprise | In-region routing on Business |
| Eden AI | No Vercel account | Provider price plus 5.5% on credits; BYOK | Cheaper providers first; fallbacks you list | Lists SOC 2 and ISO 27001; SSO on a custom plan | EU endpoint, EU models only |
| Pydantic AI Gateway | No Vercel account | BYOK at no markup, or 3 to 5% on managed providers | Provider priorities and weights you set | SSO, SCIM and audit logs on Enterprise | EU region |
| Bifrost | No Vercel account | Free (Apache 2.0); Enterprise price not published | Complexity Router tiers; fallback and weights per key | SSO, SCIM and audit logs in Enterprise | Wherever you host it |
| Opper | Hosted in the EU | Provider rates plus 3% on credits (Gateway plan) | Aliases and routes you set; unbilled fallbacks | ISO/IEC 27001:2022 | Hosted in Stockholm |
| orq.ai | Hosted in the EU | List price plus 4.5% on credits; BYOK free to 1M requests | Smart Router by difficulty band; fallbacks | SSO, SCIM and audit logs on Enterprise | EU-only platform |
| nexos.ai | Hosted in the EU | Provider price plus 5% | Smart router; budgets before overruns | SOC 2 Type 2, ISO 27001, ISO 42001; SSO on Enterprise | Hosted in Europe |
| Cortecs | Hosted in the EU | Provider price plus 5% | Ranks EU providers by speed or cost; fallback | ISO 27001; SSO not in its docs | EU providers only |
| EUrouter | Hosted in the EU | Free with 15% markup to €99 a month with 3% | Provider choice by health and price; fallback | Published DPA; no SOC 2 or ISO 27001 claim found | Processed in the EU |
| Portkey (Prisma AIRS AI Gateway) | Enterprise and cloud | Free, or $49 a month for 100k logs; your providers bill tokens | Conditional routing and fallbacks you write | SSO and SCIM on Enterprise; SOC 2, ISO 27001, HIPAA | VPC data plane on Enterprise |
| Kong AI Gateway | Enterprise and cloud | From $25 a month plus usage and $100 per model; BYOK | Cost, latency, usage or semantic load balancing; fallback | SSO and audit logging on Enterprise | Konnect EU region; self-managed data planes |
| TrueFoundry | Enterprise and cloud | Free for 3 users; Pro $25 per user a month; you pay providers | Fallback; weighted and latency load balancing | Lists SOC 2, HIPAA and GDPR | Data residency on Enterprise |
| Azure AI Foundry | Enterprise and cloud | Per token at Azure prices | Model router: balanced, cost or quality; failover | Entra ID and Azure RBAC | Data Zone EU, not for Claude |
| Amazon Bedrock | Enterprise and cloud | Per token at AWS prices | Prompt routing within one model family | AWS IAM | EU inference profiles |
| Vercel AI Gateway, for reference | List-price tokens on Vercel | Provider list price, no markup; card fees | Provider order you set; classifier left to you | SAML SSO on Vercel Enterprise | EU inference per request |
From each vendor's own pricing, docs and GitHub pages, checked October 13 and 14, 2026; "not published" means we found no answer there. Standard Compute is our product.
Cost routing built in
Through any zero-markup gateway a frontier token costs the same; the lever is which model handles each request. Vercel picks a provider for the model you name, and its cost-routing guide leaves the classifier to you. These four pick the model for each request themselves.
1. Standard Compute
A hosted gateway that optimizes for cost. With model standardcompute, the router sends each step to the cheapest model that can do it, with no classifier to build, and for every model we vet providers for price, uptime and speed, falling back to another when one fails. One key serves OpenAI-compatible agents, Claude Code (/v1/messages) and Codex (/v1/responses). Opt-in Compliance rules keep a company, team or key on provider endpoints in Europe or the US; a team or key can narrow the company rule, never widen it. We don't train on your prompts or store them. Each key runs on a fixed monthly plan with the compute included; new requests get HTTP 402 at the budget, with no overage.
Wrong pick when you pay per token and want list price with zero markup (our pay-as-you-go top-ups are usage plus 5%), you want Claude Code on your Claude Max subscription, or you route through your own provider keys (we don't support BYOK). Also when procurement needs a SOC 2 report or self-serve SCIM: we have no SOC 2 report yet, and SCIM is on request (SSO, a signed DPA and an uptime SLA come with Enterprise or as add-ons).
Vercel AI Gateway vs Standard Compute, side by sideProduction gateways comparedFor production agents
2. Kilo Gateway
Passes tokens through with no markup (a 5% fee applies to payments). Kilo Pass turns a monthly payment into credits, with bonus credits that grow with a streak. Its kilo-auto/efficient router classifies each request's difficulty and picks the cheapest model it considers accurate enough, the same idea Standard Compute uses.
Wrong pick when you run Claude Code or Codex, whose Messages and Responses APIs its gateway docs do not cover, or you need EU inference below Enterprise.
3. Cloudflare AI Gateway
A free gateway layer on any Cloudflare account, with caching, analytics, rate limits and spend limits that can block or fall back to a cheaper model. Pay providers with your own keys or with Unified Billing credits. Its Auto Router (cloudflare/auto) classifies each request's type and difficulty and weighs quality against cost, and it has guides for Claude Code, Codex, OpenCode and Pi.
Wrong pick when you want Claude Code auto-routed (the Auto Router takes Chat Completions and Responses requests, not Anthropic Messages), Codex with non-OpenAI models, or a hard cap: spend limits are recorded after each request, so bursts can pass them.
Cloudflare AI Gateway alternativeCloudflare AI Gateway vs OpenRouter
4. CodeRouter
The closest product to ours: a monthly plan for coding agents that classifies each request by phase (plan, implement, debug, small edit) and picks from 25 Claude, GPT, Gemini and open models, with Anthropic Messages and Responses endpoints for Claude Code and Codex. Plans are counted in tokens, with a separate, smaller Opus allowance.
Wrong pick when you want to pin a model on a plan (its smart-routing plans rewrite model names to auto), you don't want overage billed automatically up to a monthly cap, or you need region selection below Enterprise.
Spend limits that turn requests away
Vercel's own docs call its budgets a soft cap, and spend on your own provider keys isn't counted by default. By their own docs, each of these has limits per key that refuse requests once reached; Requesty answers HTTP 412 at a key's cap, and LiteLLM checks the budget before the provider call. Standard Compute, above, refuses new requests once the plan budget is used.
5. Requesty
One key for 600+ models at provider prices plus a fee on top-ups, or your own provider keys. Spend caps per key, fallback chains you configure, an EU endpoint in Frankfurt on every plan with an org-wide rule that rejects other regions, and native Messages and Responses support for Claude Code, Codex and OpenCode. It stores prompt content only for keys with logging on.
Wrong pick when you want the gateway to pick cheaper models for you: its routing is fallback chains, load balancing and latency policies you configure. SSO and audit logs are Enterprise features.
6. TrustedRouter
A pay-per-token router that runs its gateway in attested confidential-computing environments and keeps no prompt or output logs. Aliases such as trustedrouter/cheap, /zdr and /eu pick from fixed provider lists, and privacy minimums fail closed instead of downgrading. Per-key daily, weekly and monthly limits are built in, and its docs cover Claude Code and Codex.
Wrong pick when procurement needs a SOC 2 report (its Type II observation window is still open), or you need strict EU residency: its own EU page says upstream providers can process outside the EU.
7. LiteLLM
The most widely used open-source gateway. It speaks the Messages and Responses APIs, supports hard budgets per virtual key, team and user, and charges no fee; you pay every provider separately and can use existing cloud commitments. Its Auto Router add-on routes each request by difficulty tier, on the proxy you run, and can fill the tiers from the models you already serve. It is SOC 2 Type II audited, SSO is free for up to 5 users, and Enterprise is an annual license, never priced per token, that adds wider SSO, audit logs and air-gapped deployment.
Wrong pick when you do not want to run, patch and secure a proxy and its Postgres database. In March 2026 two LiteLLM releases on PyPI shipped a credential stealer.
8. Agentgateway
An Apache 2.0 Linux Foundation project, created by Solo.io, that proxies LLM, MCP and agent-to-agent traffic. Virtual models give weighted, priority-failover and conditional routing, load-balanced by health and latency; per-key USD budgets are true spend caps; and native Messages and Responses support backs its Claude Code and Codex guides, including running Claude Code on non-Anthropic models.
Wrong pick when you want automatic model choice out of the box (its cost routing example adds an external semantic router), or admin RBAC and audit logs as a product, which we did not find in its docs. It is self-hosted only.
Not tied to a Vercel account
AI Gateway works from any host, but every request needs Vercel authentication: keys are created in a Vercel team, and Vercel deactivates a member's keys when they leave it. These are standalone: three hosted gateways with their own accounts, and Bifrost, which you run yourself.
9. OpenRouter
The best-known pay-per-token aggregator: 500+ models behind one OpenAI-compatible API, with provider fallback, its Auto and Pareto routers, and native Messages and Responses APIs for Claude Code and Codex. You buy credits at provider price plus a fee, or bring your own keys, and prompts are not stored unless you opt in. Stripe agreed to acquire it in August 2026.
Wrong pick when you need SSO/SAML or a contractual SLA (Enterprise only), or in-region EU routing below its Business plan. Usage is billed per token, so an always-on agent's bill stays open-ended.
10. Eden AI
A French unified API for LLMs plus OCR, speech, vision and translation, at provider prices with a fee on credit purchases. When you name a model without a provider it favors the cheaper providers, falls back across providers and the backup models you list, and caches identical requests. Its EU endpoint serves only EU-compatible models and keeps prompts and outputs in the EU, its security page lists SOC 2 and ISO 27001, and Claude Code and Codex setups are documented.
Wrong pick when you need RBAC, SSO, SCIM or audit logs, which need its custom-priced Advanced AI Platform plan, or you want it to pick a cheaper model rather than a cheaper provider of the same model.
11. Pydantic AI Gateway
Now part of Pydantic Logfire: one key for many providers, with every request logged through Logfire or any OpenTelemetry backend, in a US or EU region. Bring your own keys at no markup or buy inference through it, set daily, weekly and monthly spend limits per project, user and key, and follow its documented Claude Code and Codex setups.
Wrong pick when you want the gateway to choose models: routing is priorities and weights you set per provider. SSO, SCIM and audit logs are Enterprise, and the standalone open-source gateway was archived in March 2026.
12. Bifrost
Maxim AI's Apache 2.0 gateway, written in Go, for 23+ providers on your own keys, deployed in your VPC, on-prem or air-gapped. Retries, provider fallback and weighted load balancing are set per virtual key, which also carries budgets and rate limits in the free version. Its Complexity Router sorts each request into a simple, medium or complex tier so easy work goes to a cheaper model, and it documents Claude Code and Codex.
Wrong pick when you want a hosted endpoint: we found no managed SaaS. SSO, SCIM, RBAC, audit logs and adaptive load balancing are in its Enterprise edition, whose price is not published.
Hosted in the EU
Vercel can set the inference region per request, but its gateway processing can run in any Vercel region, and team-wide provider allowlists are a paid add-on. These gateways run in Europe. Our OpenRouter alternative for EU compliance guide goes deeper on what has to stay in the EU.
13. Opper
A Stockholm-based gateway for 700+ models at provider rates, with a small fee only when you buy credits. It runs in AWS Stockholm, is ISO/IEC 27001:2022 certified, keeps only usage metadata on that plan and does not bill failed or fallback attempts. Its CLI launches Claude Code, Codex, OpenCode and other agents on the gateway in one command.
Wrong pick when you want the gateway to choose cheaper models for you, or budgets per key without moving to its Control Plane plan, where the fee is higher; on the Gateway plan, spend control is a prepaid balance.
14. orq.ai
An Amsterdam-based AI platform whose gateway runs only in the EU, at provider list prices plus a fee on its credits, or on your own keys. Its Smart Router sorts a pool of models into easy, medium and hard bands and picks by cost, balance or quality, and filters limit routing to EU inference or zero-retention providers. Budgets go down to the key or model, and Anthropic Messages and Responses endpoints back its Claude Code (beta) and Codex docs.
Wrong pick when you only need a gateway: its pricing also meters traces, data and agent runs, and audit logs, SSO and SCIM are Enterprise features.
15. nexos.ai
A European AI gateway for 200+ models at provider prices plus a platform fee, hosted in Europe and SOC 2 Type 2, ISO 27001 and ISO 42001 certified. Budgets per user, team, project or key are set before overruns happen, Claude Code has a documented setup, and the same vendor sells a chat workspace for non-technical staff.
Wrong pick when you need SSO/SAML, RBAC, audit logs or policy enforcement without an Enterprise contract, or just an API for a few developers rather than a company-wide platform.
16. Cortecs
A Vienna-based gateway that routes 150+ models across 10+ EU providers at provider price plus a service fee, ISO 27001 certified end to end. Its router filters providers by compliance, uptime and context size, ranks them by speed, cost or a balance, and retries the next one on failure. Budgets per project, team, user or key block requests at the limit, and Claude Code and Codex setups are documented.
Wrong pick when you expect zero data retention by default: it is a project setting you turn on, and without it requests can reach endpoints that keep data. Its docs don't cover SSO.
17. EUrouter
An Amsterdam-based, OpenRouter-style API with every request routed and processed inside the EU, zero data retention and a published DPA. Plans trade a monthly fee for a lower markup, and per-key spend limits return HTTP 402 at the limit.
Wrong pick when you run Claude Code: its API has no Anthropic Messages endpoint, and no Codex setup is documented. We also found no SOC 2 or ISO 27001 claim.
Enterprise platforms and your cloud
When procurement decides: SSO, audit logs and certifications from a gateway vendor, or your cloud's own model service. Our production gateway comparison lines up spend limits, prompt logging, SSO and SLAs across the main gateways.
18. Portkey (Prisma AIRS AI Gateway)
Now Palo Alto Networks' Prisma AIRS AI Gateway: a control plane for 1,600+ models on your own provider keys, priced by recorded logs. Conditional routing, load balancing and fallbacks are configs you write, Claude Code and Codex setups are documented, and Enterprise adds SSO, RBAC, VPC hosting and HIPAA.
Wrong pick when you want one bill that includes the tokens, or budget limits per key without an Enterprise contract. Product direction is now set by a security vendor.
19. Kong AI Gateway
Kong's AI layer, managed from its Konnect control plane (with an EU region) and run on serverless, Kong-hosted or your own data planes. Load balancing can follow cost, latency, usage or semantic similarity to model descriptions, with retry and fallback built in, and it has guides for Claude Code and for Codex on the Responses API. You bring your own provider keys.
Wrong pick when you are a small team: Plus is billed per control plane, per block of requests and per model proxied, and SSO and audit logging are Enterprise only.
20. TrueFoundry
An enterprise AI gateway priced per seat, with SaaS on Pro and VPC, on-prem or data residency on Enterprise. It does fallback, weighted and latency-based load balancing, budgets and rate limits per user, team or model, documents Claude Code and serves the Responses API. Its site lists SOC 2, HIPAA and GDPR.
Wrong pick when you want one bill that includes the tokens: TrueFoundry prices users and gateway requests, and model-provider charges are billed separately. Data residency needs Enterprise.
21. Azure AI Foundry
Microsoft's model service: OpenAI, DeepSeek, Grok, Mistral, Llama and Kimi models on an OpenAI-compatible v1 endpoint, and Claude, billed through Azure Marketplace, on the Anthropic Messages API, all per token at Azure prices. Its model router picks a model per prompt in balanced, cost or quality mode, with failover, and Data Zone EU deployments keep Azure-sold models inside the EU Data Boundary. Claude Code and Codex both have official setups, with Entra ID and Azure RBAC for access.
Wrong pick when you need Claude processed in the EU: Claude is offered only as Global Standard, which may run in any Azure region, or US Data Zone. Codex on Azure works with API keys only, not Entra ID.
22. Amazon Bedrock
AWS's model service, billed on your AWS account: Claude, OpenAI GPT, DeepSeek, Qwen, Kimi, GLM, Mistral, Llama and more, per token at AWS prices. EU geographic inference profiles keep requests inside the EU, Claude Code has an official Bedrock mode, Codex has a built-in Bedrock provider, and IAM controls access and cost attribution.
Wrong pick when you want one OpenAI-compatible endpoint for every model: Bedrock does not serve Claude on its Chat Completions or Responses APIs, and Intelligent Prompt Routing only chooses between two models of the same family.
How we chose
Every option is a hosted or self-hosted API that an app or agent can point at instead of Vercel AI Gateway. We started from the gateways AI assistants name when asked about Vercel AI Gateway, OpenRouter, Portkey and LiteLLM for production, EU and governance needs, then took each fact from the vendor's own pricing, docs or GitHub pages, checked October 13 and 14, 2026. Standard Compute is our product; it is listed first because this page leads with cost, and every entry, ours included, says when it is the wrong pick.
Sources: OpenRouter pricing; OpenRouter data collection; Portkey security; LiteLLM enterprise (SSO, SOC 2); Vercel pricing; Requesty pricing; Kilo pricing; CodeRouter pricing; Opper pricing; Eden AI pricing; Eden AI provider routing; Eden AI EU endpoint; Eden AI organization plans; TrustedRouter pricing; TrustedRouter EU; TrustedRouter SOC 2 readiness; TrustedRouter agent setup; Cloudflare AI Gateway pricing; Cloudflare Auto Router; LiteLLM Auto Router; LiteLLM security incident; Bifrost on GitHub; Bifrost pricing; Bifrost Complexity Router; Agentgateway virtual models; Agentgateway per-key budgets; Kong pricing; Kong AI Gateway load balancing; Portkey pricing; TrueFoundry pricing; TrueFoundry Claude Code docs; orq.ai pricing; orq.ai Smart Router; orq.ai EU regions; nexos.ai pricing; Cortecs pricing; Cortecs routing; Cortecs zero data retention; EUrouter pricing; EUrouter OpenAPI spec; Pydantic pricing; Pydantic AI Gateway docs; Azure Foundry deployment types; Azure Foundry model router; Claude Code on Microsoft Foundry; Codex on Azure; Bedrock OpenAI-compatible APIs; Bedrock geographic inference; Bedrock prompt routing; Claude Code on Amazon Bedrock
Which one fits your work
| Pick | When |
|---|---|
| Standard Compute | You want one fixed monthly price per developer key, with requests that stop at the budget and no overage |
| Through any zero-markup gateway a frontier token costs the same; the lever is which model handles each request, and model standardcompute routes routine steps to efficient models | |
| Europe or US region rules on a key, team or company without per-request add-on fees | |
| Codex via /v1/responses and OpenCode via models.dev on the same key as Claude Code | |
| Vercel AI Gateway | You pay per token and want list price with zero markup; our pay-as-you-go top-ups are usage plus 5% |
| Your app already runs on Vercel and the AI SDK | |
| Occasional use that the monthly free credit or a small balance covers | |
| You want Claude Code on your Claude Max subscription with Vercel's logs and budgets around it | |
| You need to pin exact models across many providers, with BYOK at no fee |
Switching takes one config change
Standard Compute is OpenAI-compatible, so any tool or SDK that lets you set a custom base URL migrates in minutes:
Base URL = https://api.stdcmpt.com/v1 API key = your Standard Compute key Model = standardcompute
Setup guides for OpenClaw, Hermes, OpenCode, Cursor, Cline, Aider and more are on the integrations page.
FAQ
What is the best Vercel AI Gateway alternative?
It depends on why you are leaving. For cost, a gateway that picks the cheapest capable model for each step, with the compute in one plan (Standard Compute), changes the bill the most; Kilo Gateway, Cloudflare AI Gateway and CodeRouter also choose the model for you. For spend limits that refuse requests, Requesty, TrustedRouter, LiteLLM and Agentgateway document per-key caps. For a gateway with its own account, OpenRouter, Eden AI or Pydantic AI Gateway; for one hosted in the EU, Opper, orq.ai, nexos.ai, Cortecs or EUrouter.
Is Vercel AI Gateway cheaper than Standard Compute?
Per token, yes: it charges provider list price with no markup, which our pay-as-you-go top-ups don't match. Standard Compute plans are fixed monthly prices ($19 to $249 for individual plans) with compute budgets equal to the price, stretched by routing. Light usage is cheaper metered; daily agent use is where a fixed plan caps the cost.
Can I run Claude Code through either?
Yes. Vercel's setup command points ANTHROPIC_BASE_URL at https://ai-gateway.vercel.sh/claude-code. For Standard Compute, set ANTHROPIC_BASE_URL to https://api.stdcmpt.com and use your Standard Compute key.
What happens when the money runs out?
On Vercel AI Gateway, requests draw from your credit balance; a budget returns HTTP 402 once spend passes it, and Vercel calls it a soft cap because the request that crosses it still completes. On Standard Compute, requests run until the month's spend reaches the plan budget, then the next request gets HTTP 402 until renewal, an upgrade or a top-up you choose. There are no overage charges.
What does a coding-agent month cost on Vercel AI Gateway?
Measured on our gateway, a coding agent on Claude Sonnet 5.5 for 20 typical working days costs about $71 at list price, and $579 if every day is busy, before card fees. On Standard Compute, the Standard plan is $89 with a $89 compute budget.
Does Standard Compute have SSO, SOC 2 or a DPA?
SSO and a DPA, yes; SOC 2, not yet. SSO (SAML), a signed DPA and a 99.9% uptime SLA come with Enterprise or as add-ons, the DPA is public at standardcompute.com/dpa, and SCIM is on request. We have no SOC 2 report yet and no bring-your-own-key. We also have company workspaces with roles, an audit log, a budget per key and Europe or US rules inherited from company to team to key. If procurement needs a SOC 2 report today, Portkey Enterprise, LiteLLM Enterprise, Kong or TrueFoundry fit better.
Sources: Vercel AI Gateway pricing; AI Gateway authentication; AI Gateway regional inference; Requesty error codes; LiteLLM budgets; Claude Code with AI Gateway; OpenAI Chat Completions API with AI Gateway; AI Gateway budgets; Cost-aware model routing guide