River AI has launched a paid inference service for open-weight models, starting with DeepSeek V4.1 Flash and GLM-5.3 Flash, its co-founder and chief executive Igor Babuschkin said in a post on X on 11 October 2026. He wrote that both models are “20% cheaper than the offerings from DeepSeek and ZAI” and that anyone who signs up now gets $25 in free credits.
River’s own homepage table lists DeepSeek V4.1 Flash at $0.12 for input and $0.48 for output, against $0.15 and $0.60 from DeepSeek, and GLM-5.3 Flash at $0.12 and $0.40, against $0.15 and $0.50 from Z.ai. DeepSeek’s pricing page shows that its $0.15 and $0.60 are its off-peak rates per million tokens; its peak rates are twice as high.
Business Pill · AN INTRODUCTORY PRICE
A one-minute explainer of introductory prices: a launch price or offer is low today, not necessarily forever. It teaches the general idea only and says nothing about any person or organisation in this story.
The key insight: As we read it, the number to watch is the hour, not the 20%. River’s table compares its prices with DeepSeek’s off-peak rates, and DeepSeek’s own page doubles those rates in weekday peak hours, so the gap a customer sees depends on when the tokens are bought.
River’s company account posted at 20:35 UTC that “River Inference is live”, serving DeepSeek V4.1 Flash and GLM 5.3 Flash “for 20% below official provider rates”. Babuschkin’s post followed seven minutes later.
He gave a reason for the price: “At River AI we’ve written our own training and inference stack from scratch in Rust, which unlocked much higher efficiency. We’re passing the savings on to our customers.”
The homepage presents inference next to training. It tells customers to “Use reinforcement learning to sharpen their reasoning, tool use, and ability to get things done” and to “Train with River, then serve on River Cloud or run the model on your own GPUs”. It headlines the new service as “20–60% cheaper”.
Output price for DeepSeek V4.1 Flash in US$ per 1M output tokens: River AI’s launch price against DeepSeek’s off-peak and peak rates and two other hosts, as listed on River’s homepage and DeepSeek’s API pricing page on 12 October 2026.
What Was Posted on X
River Inference is live! We’re serving DeepSeek V4.1 Flash and GLM 5.3 Flash for 20% below official provider rates.
River AI’s launch post of 11 October 2026: River Inference serves DeepSeek V4.1 Flash and GLM-5.3 Flash at 20% below official provider rates.
Weโve launched fast production inference for open-weight models! Anyone who signs up right now gets $25 in free credits to spend on the platform.
DeepSeek v4.1 Flash and GLM-5.3 Flash are 20% cheaper than the offerings from DeepSeek and ZAI.
At River AI weโve written our own training and inference stack from scratch in Rust, which unlocked much higher efficiency. Weโre passing the savings on to our customers. Enjoy!
Co-founder and chief executive Igor Babuschkin’s post minutes later, adding $25 in free credits and a Rust stack written from scratch. The piece checks the 20% claim against DeepSeek’s and Z.ai’s own price pages.
What the Prices Compare With
For DeepSeek V4.1 Flash, River’s table compares its $0.12 input, $0.0024 cached input and $0.48 output with DeepSeek at $0.15, $0.003 and $0.60, a 20% saving, and with Fireworks and Together AI at $0.30, $0.006 and $1.20, a 60% saving.
DeepSeek’s page lists those $0.15 and $0.60 figures as off-peak prices. It says off-peak rates are half of peak rates and that peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, with all other hours off-peak. At peak, DeepSeek lists $0.30 for input and $1.20 for output, so by our arithmetic River’s prices are 60% lower in those hours.
For GLM-5.3 Flash, Z.ai’s page lists $0.15 for input, $0.03 for cached input and $0.50 for output. River lists $0.12, $0.024 and $0.40, which by our arithmetic is 20% lower on each line. River’s table shows Fireworks and Together AI at the same prices as Z.ai.
GLM-5.3 Flash input and output prices in US$ per 1M tokens, Z.ai against River AI, as listed on Z.ai’s pricing page and River’s homepage on 12 October 2026.
Who Is Behind It
Babuschkin describes himself on X as chief executive and co-founder of River AI, and previously at xAI. We covered River’s funding in August, when it raised $1.1 billion to help enterprises train and own custom models.
On the new service, River sells per-token access to models that DeepSeek and Z.ai also serve themselves, and its table sets its prices against theirs and against two other hosts, Fireworks and Together AI.
The Structural Read
As we read it, River sells the same two open-weight models that DeepSeek and Z.ai serve themselves, so the comparison on its page holds the model fixed and varies the price.
As we read it, the company’s stated reason is cost: Babuschkin says the Rust stack River wrote “unlocked much higher efficiency” and that River is passing the savings on.
The launch offer of $25 in free credits applies to “Anyone who signs up right now”, in Babuschkin’s words; the post gives no end date.
Igor Babuschkin, River AI co-founder and CEO, on X, 11 October 2026
“At River AI weโve written our own training and inference stack from scratch in Rust, which unlocked much higher efficiency.”
Three Implications
DEVELOPERS ON DEEPSEEK V4.1 FLASH River lists $0.12 per million input tokens and $0.48 per million output tokens; DeepSeek lists $0.15 and $0.60 off-peak and $0.30 and $1.20 at peak.
DEVELOPERS ON GLM-5.3 FLASH River lists $0.12 for input, $0.024 for cached input and $0.40 for output, against $0.15, $0.03 and $0.50 on Z.ai’s page.
OTHER HOSTS River’s table names Fireworks and Together AI at $0.30 and $1.20 for DeepSeek V4.1 Flash and at Z.ai’s prices for GLM-5.3 Flash.
The framework puts it this way: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query”.
As we read it, River’s launch is a bet on that marginal cost: the same open models, served at a lower price per token, with the difference credited to software River wrote itself.
What Is Not Established
The posts and pages we read give no throughput, latency or uptime figures for River’s service, no customer names and no end date for the $25 offer. The efficiency gain from the Rust stack is River’s own account. The comparison with DeepSeek depends on the hour: River’s 20% figure matches DeepSeek’s off-peak rates.
Business Engineer Framework
The State of The Inference Economy
A Business Engineer framework on why inference, not training, is now where AI money is made and spent.
River AI has started selling inference for two open-weight models at prices its own table sets 20% below DeepSeek’s off-peak rates and Z.ai’s list rates. Babuschkin credits a stack River wrote from scratch in Rust, and new users get $25 in credits.
A note on sourcing. On 12 October 2026 we read Igor Babuschkin’s and River AI’s posts on X of 11 October (through a mirror of X’s public data), Babuschkin’s X profile, the price tables on River’s homepage, DeepSeek’s API pricing page and Z.ai’s pricing page. Every price here comes from those pages; the claims about cost savings and River’s Rust stack are the company’s own. Percentage differences labelled as ours are our arithmetic. Nothing here is a forecast or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
