Pricing

Simple, per-user pricing

Prices are shown per month. Start free with a few coworkers, add a usage budget when you need more, and move to dedicated infrastructure when your organization does.

Playground

Always free

$0/ user / month

✓A free workspace to build multiplayer bots

Team

Starts at

$40/ paid user / month

Every user starts free. Pay only for users who need more in $40 increments, billed weekly.

✓Shared context: Knowledge, Skills, Semantic layer

Enterprise

Starts at

$1,000/ month

Consumption based with a monthly minimum commit

✓No limit on users

✓Shared context: Knowledge, Skills, Semantic layer

✓Single-tenant deployment

Custom

Priced to fit

Talk to us

Contract pricing

✓BYOC or self-hosted deployment

✓Bring your own models

✓Dedicated forward-deployed engineers

Your model choice changes your mileage.

OLUs are the normalized unit we use to measure usage across models. Frontier models like Claude Fable 5 and GPT-6 Astra consume the most OLUs; open-weight models like Kimi K3, DeepSeek V4 and GLM 5.2 do the same work for a fraction of the OLUs. Switch to different models for different bots — or even mid-thread.

Model	Type	OLU multiplier (vs Opus 4.6)	Input tokens / OLU	Output tokens / OLU	Cached tokens / OLU	Cache creation tokens / OLU	Notes
Open-weight models · served via Fireworks
DeepSeek V4 Flash 0731 Fireworks	Open-weight	0.025x (40x cheaper)	952,400	595,200	6.0M	1.2M	Served via Fireworks. 07/31 snapshot variant of DeepSeek V4 Flash.
Llama 3.3 70B Fireworks	Open-weight	0.036x (28x cheaper)	493,800	617,300	6.2M	617,300	Meta; served via Fireworks.
DeepSeek V4.1 Flash Fireworks	Open-weight	0.04x (25x cheaper)	606,100	252,500	23.8M	757,600	Multimodal DeepSeek model with native vision; served via Fireworks. $0.22/$0.007/$0.66 per 1M in/cached/out.
Qwen 3.7 Plus Fireworks	Open-weight	0.071x (14x cheaper)	333,300	104,200	4.2M	416,700	Alibaba, vision-capable; served via Fireworks.
Mistral Large 3 Fireworks	Open-weight	0.077x (13x cheaper)	333,300	83,300	4.2M	416,700	Mistral AI; served via Fireworks.
Kimi K2.7 Fireworks	Open-weight	0.17x (5.8x cheaper)	140,400	41,700	1.8M	175,400	Moonshot AI 1T-param MoE; served via Fireworks.
GLM-5.2 Fireworks	Open-weight	0.23x (4.3x cheaper)	95,200	37,900	1.2M	119,000	Z.ai (Zhipu), MIT-licensed; served via Fireworks.
DeepSeek V4 Pro 0813 Fireworks	Open-weight	0.24x (4.2x cheaper)	101,000	42,100	3.6M	126,300	Official GA snapshot (0813) of DeepSeek V4 Pro; served via Fireworks. Supersedes the preview. $1.32/$0.044/$3.96 per 1M in/cached/out.
DeepSeek V4 Pro (Fireworks) Fireworks	Open-weight	0.25x (4x cheaper)	76,600	47,900	1.1M	95,800	Preview (deepseek-v4-pro), still billed at 0.25x / $1.74/$0.145/$3.48 per 1M in/cached/out; superseded by 0813. Served via Fireworks (what PromptQL uses for open-weight).
Kimi K3 Fireworks	Open-weight	0.6x (1.7x cheaper)	44,400	11,100	555,600	44,400	Moonshot AI flagship; served via Fireworks.
Proprietary models
GPT-5.6 Luna OpenAI	Proprietary	0.04x (25x cheaper)	133,300	33,300	1.7M	166,700	Cost-efficient GPT-5.6 tier.
Gemini 3.7 Flash Google	Proprietary	0.15x (6.7x cheaper)	177,800	44,400	2.2M	222,200	Google Gemini Flash tier; ~2x cheaper than Gemini 3.6 Flash.
Gemini 3.8 Flash Google	Proprietary	0.15x (6.7x cheaper)	177,800	44,400	2.2M	222,200	Latest Google Gemini Flash tier; same introductory pricing as Gemini 3.7 Flash through Dec 31, 2026.
Claude Haiku 4.5 Anthropic	Proprietary	0.2x (5x cheaper)	133,300	33,300	1.7M	133,300	Fast tier.
GPT-5 OpenAI	Proprietary	0.3x (3.4x cheaper)	106,700	16,700	1.3M	133,300	GPT-5 base.
GPT-5.1 OpenAI	Proprietary	0.3x (3.4x cheaper)	106,700	16,700	1.3M	133,300	Same rate as GPT-5.
Grok 4.5 SpaceXAI	Proprietary	0.37x (2.7x cheaper)	66,700	22,200	666,700		SpaceXAI (xAI) frontier coding/agentic model. No cache-creation pricing.
Grok 4.6 SpaceXAI	Proprietary	0.37x (2.7x cheaper)	66,700	22,200	666,700		SpaceXAI (xAI) frontier coding/agentic model. Same pricing as Grok 4.5. No cache-creation pricing.
Gemini 3.1 Pro Google	Proprietary	0.42x (2.4x cheaper)	66,700	13,900	833,300	83,300	Google flagship Pro.
GPT-5.2 OpenAI	Proprietary	0.42x (2.4x cheaper)	76,200	11,900	952,400	95,200	Flagship step-up.
GPT-5.6 Terra OpenAI	Proprietary	0.42x (2.4x cheaper)	53,300	11,100	666,700	66,700	Balanced GPT-5.6 tier.
GPT-5.4 OpenAI	Proprietary	0.52x (1.9x cheaper)	53,300	11,100	666,700	66,700	Reasoning/coding/agentic gains.
Claude Sonnet 4.5 Anthropic	Proprietary	0.6x (1.7x cheaper)	44,400	11,100	555,600	44,400	Balanced tier. (Sonnet 4.6 is blocklisted in PromptQL.)
Claude Opus 4.6 Anthropic	Proprietary	1.0x (baseline)	26,700	6,700	333,300	26,700	Anchor (1.0x).
Claude Opus 4.8 Anthropic	Proprietary	1.0x (baseline)	26,700	6,700	333,300	26,700	Current GA Opus; PromptQL Powerful tier. Same rate as 4.6/4.7.
Claude Opus 5 Anthropic	Proprietary	1.0x (baseline)	26,700	6,700	333,300	26,700	Latest GA Opus. Same rate as 4.8/4.6.
GPT-5.5 OpenAI	Proprietary	1.0x (baseline)	26,700	5,600	333,300	33,300	Latest flagship. >272K-context billed 2x in / 1.5x out.
GPT-5.6 Sol OpenAI	Proprietary	1.0x (baseline)	26,700	5,600	333,300	33,300	GPT-5.6 flagship. GA July 9, 2026.
Claude Fable 5 Anthropic	Proprietary	2x (2x cost)	13,300	3,300	166,700	13,300	New frontier flagship (Jun 2026); most capable, highest cost.
GPT-6 Astra OpenAI	Proprietary	2x (2x cost)	13,300	3,300	166,700	13,300	OpenAI GPT-6; $10/$50 per MTok in/out. Formula (10 + 0.05*50)/6.25 = 2.00.
GPT-5 Pro OpenAI	Proprietary	7.22x (7.2x cost)	8,900	1,400	11,100	11,100	Pro reasoning. No prompt-cache discount.
GPT-5.2 Pro OpenAI	Proprietary	10.11x (10.1x cost)	6,300	992	7,900	7,900	Pro reasoning. No prompt-cache discount.
GPT-5.4 Pro OpenAI	Proprietary	13.53x (13.5x cost)	4,400	926	5,600	5,600	Pro reasoning. No prompt-cache discount.
GPT-5.5 Pro OpenAI	Proprietary	13.53x (13.5x cost)	4,400	926	5,600	5,600	Top GPT pro tier. No prompt-cache discount.

Anchor: Claude Opus 4.6 = 1.0×. Lower multiplier = cheaper, i.e. more tokens of work per OLU (per $0.20). PromptQL serves open-weight models via Fireworks (DeepSeek is shown for both its first-party Official API and the Fireworks-hosted rate PromptQL bills on). PromptQL gives you access to every model available, and the lineup expands over time — pick your model per thread. The GPT *-pro tiers run higher mainly because they have no prompt-cache discount. Token figures are representative; your exact mileage depends on the task.

Frequently asked questions

Playground is a free workspace where you build bots and share them, within basic usage limits. On Team, every user starts free with a basic usage limit, and paid usage plans start at $40 per user per month, billed weekly as $10 per user per week. Enterprise is a $1,000 per month minimum commit with no limit on users and a single-tenant deployment. Custom covers BYOC, self-hosted and bring-your-own-model deployments.

A usage budget. Each Team user on a paid plan gets $40 of usage a month, billed as $10 each week, and you can assign a higher plan to anyone who needs it in $40 per month increments. The budget is spent as that user's bots consume OLUs. Invoices are weekly, and unused credits roll over to the next week.

An Operational Language Unit (OLU) is PromptQL's normalized unit for AI usage. It combines input, output and cached tokens, then applies the selected model's multiplier, so usage across different token types and models can be measured in one unit. More complex or longer work generally consumes more OLUs.

On Team, usage is billed at cost with no margin during the introductory period: $10 buys $10 worth of tokens, VM runtime and any other billable usage. Enterprise starts at $1,000 per month, fully allocated to usage, with OLU rates set by contract.

Yes. Your price per OLU is fixed by your plan or contract, while the model multiplier changes how many OLUs the model's token usage consumes. A 0.20× model uses one-fifth the OLUs of a 1.0× model for the same token mix. The model → OLU table above shows the multiplier and representative tokens per OLU for every model.

How much context a bot reads, how much it writes, the mix of input, output and cached tokens, and the model you choose. You control those dials by picking the right model and keeping each bot focused on the context it needs. A running VM also consumes OLUs, as do built-in integrations like Exa, Gemini or XAI.

All model, infrastructure and sandbox-hosting charges. That covers the LLM tokens plus the orchestration and the isolated computer each bot runs on: VM, files, artifacts and schedules. There is no separate LLM pass-through or infrastructure bill.

Yes. Connect your own OpenAI Codex subscription under bring-your-own-model and PromptQL charges $0 for usage on that connected model. Usage on every other model is deducted from your OLU budget as usual.

Yes. On Team, every user starts on the free tier with a basic usage limit, and you can move anyone to a paid plan from $40 a month ($10 a week), in $40 per month increments. When a user reaches their limit, their bots pause until the budget resets or you raise the plan. Enterprise sets per-user weekly limits with no cap on the number of users.

Start on Free: create a project, invite coworkers, connect your data and start a bot. Upgrade to Team when you need a bigger usage budget, or talk to us about Enterprise when you need dedicated infrastructure and organization-wide controls.