Neuralwatt Doubling the Price

wpnews.pro

Energy-based pricing aligns your costs with actual GPU resources consumed.

New pricing rolling out July 16 #

We've been thrilled by the response to Neuralwatt Cloud and the launch of flex inference. To support our growing scale and invest in reliability, capacity, and new pricing tiers, we're updating our standard energy rate from $5/kWh to $10/kWh on July 16, 2026. Energy pricing will still be substantially cheaper than token-based pricing for most workloads, and the efficiency compounds with flex.

What's changing July 16

Base energy rate: $5 → $10/kWh - PAYG packs become flexible: buy any amount from $5–$1000 at the flat rate

Subscription tiers adjusted for the new rate

Temporary (until July 16)

$1,000 PAYG packs andannual subscriptions are temporarily unavailable while we prepare the new pricing - Monthly subscriptions, $10–$50 packs, and all existing credits remain fully available

Your current credits and subscriptions are unaffected

What's coming: pricing tiers by flexibility

We're building tiered pricing that lets you (and your agents) optimize price vs. performance depending on how much latency you can tolerate. The following are early examples. Exact discount levels and availability are still being finalized.

Latency-tolerant inference at a discount. Greater hold tolerance = deeper savings.

Synchronous inference with improved SLA and throughput guarantees.

Priority lanes with tighter TTFT and throughput SLAs, billed at a premium.

Dedicated capacity, sovereign data residency, higher concurrency, tighter SLAs.

We're also working on API access to real-time energy/request data, flex hold times per discount level, and throughput metrics, so your agents can self-optimize price/performance. More detail to come as these ship.

We're grateful for your engagement and support as we grow. If you have questions, reach out in our Discord or email

[email protected].

Monthly Subscription Plans #

Predictable monthly pricing with included energy allocation. Know exactly what you'll pay each month.

Basic

Perfect for personal projects and light AI workloads.

Access to all models
Real-time energy tracking
Energy dashboard

Standard

Great value for everyday AI productivity with priority access.

Everything in Basic
2.7x more energy than Basic
Priority access during high demand
Usage analytics

Pro

For power users with the most energy and highest priority.

Everything in Standard
5.5x more energy than Basic
Highest priority access
Overage billed at your plan's discounted rate (once your plan renews after July 16)

* As part of the [July 16 pricing update](#pricing-update),
new allowances reflect the $10/kWh rate with your tier's discount (2% Basic, 3% Standard, 5% Pro) and apply to

subscriptions that start or renew on or after July 16. A subscription bought today keeps its current allowance for its full term; nothing changes mid-cycle.

Why Subscribe?

Subscriptions are billed in kWh (energy), not tokens. This means the same rate applies to all models: use a small, fast model or a large reasoning model with zero markup. You always know exactly what you'll pay.

API Access & Pay As You Go #

No commitment required. Start with $1.00 free credits, then add more when you need them.

Energy-Based Pricing

Pay for actual compute consumed

Pay for actual GPU compute consumed, not arbitrary per-token markups. One rate for all models: efficient MoE models cost dramatically less than traditional token pricing.

- One flat rate, no per-model markups
- MoE models (Qwen3.5, Kimi K2.5) up to 95% cheaper

Efficient prompts cost less, so optimization pays off
Full cost transparency
Prefix caching: cached tokens skip recomputation, reducing energy vs full prefill

Token-Based Pricing

Alternative for familiar billing

Standard per-token billing if you prefer familiar pricing. Rates vary by model; consider energy pricing for better value on most models.

Familiar per-token model
Easy migration from other providers
Energy reporting included free

Free Trial

$1.00 credit included, 5 req/min, 50 req/day

Pay As You Go

$10 minimum, 500 req/min, unlimited daily

Model Pricing

Token prices vary by model. Energy pricing uses a flat $5.00/kWh for all models, dramatically cheaper for efficient models. The energy figure below is the average measured energy at a typical request size from real usage over the trailing 7 days, not a fixed rate; energy scales with request size.

Understand energy pricing & compare costs → | Model | Input | Cached Input† | Output | Avg Energy/Req | | |---|---|---|---|---|---| | GLM-5.2 ZhipuAI | $1.45/M tokens | $0.36/M tokens | $4.50/M tokens | ~1.97 Wh | |

Try Try Try Try Try## Team & Enterprise Dedicated infrastructure, custom SLAs, and volume discounts for organizations.

Enterprise

For organizations that need dedicated GPU infrastructure, guaranteed capacity, and enterprise-grade support.

[Contact Sales](/cdn-cgi/l/email-protection#771e1911183719120205161b001603035914181a)

Neuralwatt Deploy

Bring Neuralwatt's energy optimization directly into your data center. The same Neuralwatt Optimize engine powering Neuralwatt Cloud, deployed on your infrastructure.

Full control over your hardware, security, and power consumption. We'll help you deploy the same optimizations powering this service.

All Models #

Token-based pricing varies by model. Energy-based pricing uses a flat $5.00/kWh rate for all models, up to 95% cheaper on efficient architectures.

How we measure energy †: $0.36/M tokens

Try Now †: $0.36/M tokens

Try Now †: $0.17/M tokens

Try Now †: $0.24/M tokens

Try Now †: $0.17/M tokens

Try Now †: $0.07/M tokens

Try Now | Model | Context | Token Input | Cached Input† | Token Output | Avg Energy/Req* | Features | | |---|---|---|---|---|---|---|---| | GLM-5.2 ZhipuAI | 1048K | $1.45 /M tokens | $0.36 /M tokens | $4.50 /M tokens | ~1.97 Wh | R T | |

Try Try Try Try Try Try Try Try Try Try *Average measured energy at a typical request size from real usage over the trailing 7 days, pooled across all servers, not a fixed rate. Energy scales with request size; see the full grid.

Learn more †Cached input tokens are billed at 25% of the input rate. Prefix caching makes long-context requests substantially cheaper when the same prefix is reused. The 25% is the default rate; per-model rates may be set in future, in which case the per-model rate applies. This same effective rate is advertised as cached_input_per_million

on /v1/models.

Frequently Asked Questions #

Energy pricing is recommended for most users; it's dramatically cheaper for efficient MoE models (up to 95% savings vs token pricing) and gives you full cost transparency.

Subscriptions combine energy pricing with a monthly discount (~35% off PAYG rates).

Token pricing is available as a familiar alternative if you prefer per-token billing. All options include free energy reporting.

source & further reading

portal.neuralwatt.com — original article