Agent pricing takes shape: usage models, factory tiers and platform fees replace flat plans
Twilio moved Video to pay-as-you-go at $0.004 a minute, X added Imagine image and video models, Warp launched five Factories tiers with a $10,000 Early Access credit, MindsDB added a 5% platform fee and DeepSeek exempted weekends from peak pricing. 209 changes across 206 companies.
This week in pricing · Aug 17 – Aug 23
2026W35209
+0% vs 8-wk avgpricing changes across 4,388 tracked pricing pages · 206 companies







The moves
Twilio overhauled Video pricing, dropping the 5,000 free hours plus $1 per active-user-hour or $150 per named user per month structure in favor of pure pay-as-you-go billing at $0.004 per participant per minute [13]. Pricing comparisons describe Twilio's two video products, P2P for small rooms and Video Groups for larger ones [1]; two weeks later our capture logged the hours model returning [2], so this is the first half of a round trip.
X narrowed the main pricing table to grok-4.6, with grok-4.5 only referenced in a collapsed FAQ, and added two Imagine models, grok-imagine-image-2.0 from $0.04 per image and grok-imagine-video-1.5 from $0.08 per second; the Build on your own plan now points at a free Playground for testing before paying [14]. Routers list Image 2.0 from $0.04 per image at low 1K to $0.08 at medium 2K, with $0.01 per image input [3], and product guides name Image 2.0 and Video 1.5 as the current API recommendations [4].
Warp added a Warp Factories pricing section with five tiers, Pay as you go, Build, Max, Business and Enterprise, with pay as you go billed at a 20% markup and up to $10,000 in factory usage for select teams during Early Access, and removed the Claude Fable 5 and Opus 5 50% off promo banner from the top of the page [15]. The launch release describes the product as infrastructure and control plane with teams customizing their own factories [5]; coverage notes it is closed beta only, with access by application [6], and that it works with any model and any harness [7].
MindsDB split MindsHub's Free and Pro plans into Cowork plus a new Unified Inference API product, both adding a 5% platform fee on rate-card prices; Gemini 3.6 Flash rates were halved and Kimi K3 gained a cache-write fee [16]. The pricing page today describes MindsHub as Agents plus Inference on one rate card [8][9], so the naming has moved again since the capture.
DeepSeek changed its peak-hours footnote so peak pricing applies Monday through Friday only, making weekends always off-peak, and added deepseek-v4-flash-vision-exp to the table at the same price as deepseek-v4-flash, $0.22 off-peak and $0.44 peak per 1M cache-miss input tokens and $0.66 off-peak and $1.32 peak per 1M output tokens, with images billed as input tokens [17]. The pricing page now states peak hours as 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, excluding Chinese public holidays [10], with all other hours off-peak including weekends [11]; coverage confirms off-peak is half of peak in every row [12].
What they said
Warp framed Factories as infrastructure for cloud software factories, with teams retaining the flexibility to customize and extend them [5]. DeepSeek stated the schedule plainly in the footnote: weekdays peak, everything else off-peak [10][11]. MindsDB describes one rate card shared by its agent workspace and its API [8]. Twilio and X changed their pages without an announcement our capture could point at [13][14].
What the market said
Coverage followed the model launches, not the pricing mechanics. Grok Imagine write-ups carry the per-image and per-second rates [3][4]. Warp Factories coverage focused on the closed beta and the bring-your-own-harness design [6][7], not the 20% pay-as-you-go markup. On DeepSeek, explainers had already worked out that off-peak is half of peak and that peak follows the Beijing workday [12]. Twilio's switch to per-minute billing drew no coverage before it was reversed [1][2].
The pattern
Pricing pages logged 209 changes across 206 companies in the week of Aug 17 → Aug 24, a 1.0 ratio to the eight-week average of 199.1 [18]. Packaging dominated: 101 packaging changes against 51 pure pricing changes and 57 product changes, with 88 expansions at a 1.2 ratio and 32 contractions [18].
Price direction was even: 17 increases, a 1.0 ratio to the mean of 16.9, against 8 cuts, with 3 hidden-pricing events and 7 reveals [19]. Promotions were below normal at 16, a 0.8 ratio to the average of 19.3 [20], and Warp removed one of them [15].
The through-line was agent pricing taking shape: usage-based video, per-image and per-second generation, factory tiers with markups, a platform fee on inference, and a peak schedule that now spares weekends. Flat plans gave way to meters almost everywhere the week touched.
What to watch
Whether Twilio keeps per-minute Video billing or returns to free hours (it returned two weeks later) [2].
Whether X publishes resolution tiers for Imagine Video beyond the from $0.08 per second entry rate [3].
Whether Warp opens Factories beyond the closed beta and keeps the 20% pay-as-you-go markup [6].
Whether MindsDB keeps the 5% platform fee as MindsHub renames its products again [8].
Whether DeepSeek's weekend exemption shifts batch workloads off weekday peaks [10].
Written with 90% of the week’s diffs published; late-published changes appear in the feed.
Sources
- [1]100ms — Navigating the Video Call API Landscape: Pricingverified“In this section, we’ll be examining the pricing charged by two of Twilio’s video products: Twilio P2P and Twilio Video Groups.”
- [2]Pulse capture (2026W37): Twilio Video ditches per-minute billing for 5,000 free hours, then $1/user-hour“Twilio Video ditches per-minute billing for 5,000 free hours, then $1/user-hour”
- [3]OpenRouter — Grok Imagine Image 2.0verified“Grok Imagine Image 2.0 costs from $0.04/image at low/1K to $0.08/image at medium/2K for Image Output and $0.01/image for Image Input.”
- [4]The Rundown — Grok Imagine: Image & Video Features, Pricing & Limitsverified“The current flagship API recommendations are Grok Imagine Image 2.0 for images and Grok Imagine Video 1.5 for video, while the consumer experience packages the tools behind modes and a shared Grok usage allowance.”
- [5]Warp press release — Warp Launches Warp Factories (Fortune / EZ Newswire)verified“Warp provides the infrastructure and control plane while teams retain the flexibility to customize and extend their factories.”
- [6]It's FOSS — Warp Launches Factoriesverified“If you or your organization wants to try it out, then you will need to request access through Warp Factories' early access portal.”
- [7]It's FOSS — Warp Factories (models and harnesses)verified“Factories works with any model, open-weight ones too, and any harness , so Claude Code or Codex can sit in the implementation seat instead of Warp's native agent .”
- [8]MindsHub — pricingverified“MindsHub Agents and MindsHub Inference use the same rate card.”
- [9]MindsHub — pricing (products)verified“MindsHub includes MindsHub Agents, the agent workspace, and MindsHub Inference, the API.”
- [10]DeepSeek API docs — Models & Pricingverified“Peak hours are 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday, excluding Chinese public holidays.”
- [11]DeepSeek API docs — off-peak hoursverified“All other hours are off-peak, including weekends and Chinese public holidays in full.”
- [12]Zentor — DeepSeek API Pricing Doubles 35 Hours a Weekverified“Off-peak rates are half of peak, in every row, with no exceptions.”
- [13]Pulse capture: Twilio ditches per-user Video pricing for pay-as-you-go at $0.004/min“Twilio ditches per-user Video pricing for pay-as-you-go at $0.004/min”
- [14]Pulse capture: X drops grok-4.5, adds Imagine image/video models and free Playground perk“X drops grok-4.5, adds Imagine image/video models and free Playground perk”
- [15]Pulse capture: Warp launches Warp Factories with 5 tiers and $10K Early Access credit, drops Claude 50% off promo“Warp launches Warp Factories with 5 tiers and $10K Early Access credit, drops Claude 50% off promo”
- [16]Pulse capture: MindsDB splits Free/Pro into Cowork + Inference API, adds 5% platform fee“MindsDB splits Free/Pro into Cowork + Inference API, adds 5% platform fee”
- [17]Pulse capture: DeepSeek adds vision model, exempts weekends from peak pricing“DeepSeek adds vision model, exempts weekends from peak pricing”
- [18]Pulse weekly stats, 2026W35Pulse Index data“209 changes across 206 companies; 101 packaging, 51 pricing, 57 product; 88 expansions (ratio 1.2), 32 contractions; eight-week mean 199.1, ratio 1.0”
- [19]Pulse weekly stats, 2026W35Pulse Index data“17 price increases (ratio 1.0 to the 16.9 mean), 8 price cuts, 3 hid pricing, 7 revealed pricing”
- [20]Pulse weekly stats, 2026W35Pulse Index data“16 promo events, ratio 0.8 to the 19.3 mean”
Questions about this week?
The Pulse Agent has every change from this issue loaded — the story, the numbers and each company's diff — and answers from verified data.
Ask the agentBy the numbers
- pricing changes tracked· about normal
- 209
- price increases vs cuts
- 17 ↑ / 8 ↓
- expanded vs contracted plans
- 88 / 32
- packaging moves
- 101
- discount & promo moves· below normal (−20%)
- 16
Previous issues
All weekly issues →- Week 41, 2026·Sep 28 – Oct 4·279 changesIntercom tests a $19 seat while the rest of the market shows its prices: 15 reveals in a week of quiet cuts
- Week 40, 2026·Sep 21 – Sep 27·276 changesPackaging rewrites outpace price moves: credits, bundles and metered add-ons
- Week 39, 2026·Sep 14 – Sep 20·223 changesSupport tiers, disclosed rates and a hidden price: packaging moved more than prices
- Week 38, 2026·Sep 7 – Sep 13·196 changesCredits replace rate limits, flat rates replace meters: a normal week with structural moves
- Week 37, 2026·Aug 31 – Sep 6·204 changesMetrics and packages rewritten: steps, hosts, hours and regions outnumber price hikes
- Week 36, 2026·Aug 24 – Aug 30·149 changesA quiet week for list prices, a loud one for metering: tokens, seats, events and parsing