What it is
Bland is a voice-agent platform for complex, high-volume phone calls: inbound support, outbound campaigns, intake and verification, sold mostly to enterprises that want the whole stack under one vendor. Its distinguishing bet is infrastructure. Bland says speech recognition, the language model and speech synthesis all run on its own servers, with no hand-off to a third-party model or voice vendor, which is what lets it offer VPC deployment on AWS, GCP or Azure and fully on-premises installs for the most sensitive workloads. That puts it between the build-it-yourself layer of Vapi and Retell AI and the managed-service layer of PolyAI and Parloa: self-serve APIs, plus forward-deployed engineers on Enterprise.
The San Francisco company, led by co-founder and CEO Isaiah Granet and backed by Y Combinator, raised a $50M Series C led by Dell Technologies Capital on 16 June 2026, with HubSpot Ventures joining; total funding passed $100M. Fortune reported that the company had been turned down by 180 investors before the round. The company reports more than 3.5 million calls a week and 250+ enterprise customers, naming Samsara, Kin Insurance and CNO Financial Group. Those volume figures are vendor-reported.
Why it shows up in CX and support stacks
- One vendor for the whole speech loop. Transcription, model and voice are billed in a single talk-time rate. There is no separate LLM line to forecast, which is the main complaint about assembling a stack on Vapi.
- Norm builds the first draft. Launched in March 2026, Norm turns a plain-English description into an agent with its prompt, conversation flow, validation conditions and extraction rules, then runs agent-on-agent simulations against a synthetic caller before you deploy. Bland positions it as the codified version of what its forward-deployed engineers used to do by hand. Norm is billed separately on tokens, at a rate the billing docs do not publish.
- Regulated buyers have a real path. Bland holds audited SOC 2 Type I and Type II reports, ISO 27001, a HIPAA BAA and GDPR DPA, self-attests to PCI DSS, and in August 2026 announced FedRAMP 20x Class A certification for its voice, SMS and chat agents. The BAA, SSO, data residency and on-prem/VPC deployment are Enterprise-only.
- Its own voices. Bland Speech v3, released in August 2026, is the current voice model; Bland’s own benchmark puts it second only to human recordings for realism. No independent benchmark confirms that ranking.
- Developer surface. A Calls API, a TTS API, a CLI, and an MCP server (
npx bland-cli mcp) that lets Claude or another client create agents and place calls. Bring-your-own Twilio or any SIP trunk is supported.
Pricing reality
Bland replaced its old flat $0.09/min rate on 5 December 2025 with a plan ladder. Start has no platform fee and bills $0.14 per connected minute, capped at 10 concurrent calls and 100 calls a day. Build is $299 a month plus $0.12/min, with 50 concurrent calls and 2,000 calls a day. Scale, listed in the billing docs but not on the public pricing page, is $499 a month plus $0.11/min, 100 concurrent calls and 5,000 calls a day. Enterprise is custom and is where uncapped volume, dedicated orchestration servers and the compliance paperwork live.
The per-minute rate is close to all-in, but three meters sit around it. Each outbound attempt on Bland telephony carries a $0.015 minimum, answered or not. Time after a transfer to a human on a Bland number bills at $0.03–$0.05/min depending on plan, and is free if you bring your own carrier. SMS is $0.02 per message.
At 100,000 minutes a month, Build comes to about $12,300/month and Scale to about $11,500, or roughly $140K–$150K a year before transfers and attempts. An outbound campaign dialing 100,000 numbers adds $1,500 in attempt fees on its own. That is in the same band as a mainstream Retell build (about $0.15/min) and overlaps a mainstream Vapi build ($0.09–$0.14/min all-in) before Vapi’s $999/month Pro floor.
Best for
CX-ops, support and RevOps leaders at mid-market and enterprise companies running regulated or high-volume phone work — insurance intake, healthcare scheduling, government services, outbound qualification at 100,000+ minutes a month — who want one contract covering speech, model and voice, and who need VPC, on-prem or FedRAMP as a hard requirement rather than a nice-to-have.
Skip it if your team wants to choose its own LLM and voice per agent, if you are below a few thousand minutes a month, or if you need a HIPAA BAA without an Enterprise contract.
Versus the alternatives
Vapi is the developer default by reach, with over 1 billion calls handled and a $50M Series B in May 2026. Pick Vapi when engineers want to swap every component; pick Bland when the swap is the problem and you want one meter. Retell AI is the other high-volume default; pick it when a small team needs a self-signed BAA and built-in QA on every call. ElevenLabs is the fastest-growing entrant, at $500M+ ARR; pick it when voice quality and language coverage decide the project. Synthflow is the pick when the builder must be a non-engineer. If you want a vendor to design and run the agent, buy managed: PolyAI, Parloa or Sierra. If your contact centre already runs Genesys Cloud, NICE CXone or Five9, test their native AI agents against your containment bar first.
If none fit, put your single highest-volume intent on Bland’s free Start tier and on one alternative for two weeks, and compare measured cost per resolved call, not list price.
Watch-outs
- A forced builder migration lands on 15 November 2026. Bland’s docs say production pathways and personas must move to the new Agents architecture by that date, and there is no one-click import from other platforms. Guard: export each pathway’s JSON through the API, migrate a non-production copy first, and re-run your simulation suite before switching traffic.
- Performance and realism claims are self-published. The 200ms response time and the realism benchmark come from Bland’s own site. Guard: measure end-to-end latency and caller drop-off in your pilot, on your own numbers, and write a latency threshold into the Enterprise order form.
- The low tiers throttle outbound. Start allows 100 calls a day and Build 2,000, and failed dials still cost $0.015 each. Guard: size daily attempts and peak concurrency from your dialer data before choosing a plan, and clean lists for disconnected numbers before a campaign.
- Compliance sits behind Enterprise. The BAA, SSO and data residency are not available on Build or Scale. Guard: if a call touches PHI or EU personal data, price Enterprise from day one rather than piloting on a self-serve tier you cannot put into production.
- Owning the whole stack means owning its outages. With no third-party model to fail over to inside the platform, a Bland incident is a full outage. Guard: keep a human fallback number in your IVR and run telephony on your own Twilio or SIP trunk so numbers move with you.