InvestNot yet confirmed elsewhere1 publisher2 min readPublished
DeepSeek V4 doubled its OpenRouter token share, and the bill it displaced was ~130x bigger
OpenRouter says DeepSeek went from 9% to 18% of tokens in six months, with agent loops doing most of the work. Token share and spend share are not the same league table.
The Investor · Invest desk
What happened
- Before V4, DeepSeek had slipped to 5% of platform tokens as agentic usage took off in February and March.
- V4 Flash accounted for 70% of DeepSeek's agentic token flow by the end of May, a month after launch.
- Chinese-authored models passed American ones in OpenRouter token share in early June, on OpenRouter's numbers.
Why it matters
- cost The cheap endpoint is what makes loop-heavy work payable, and every agent run left on a frontier endpoint now carries a premium that has to be argued for on quality rather than inherited by default.
- exposure The incumbents are losing the volume tier first, which is exactly where loops accumulate; revenue can look intact for a while as the token footprint erodes underneath it.
- constraint Leading on tokens while lagging on spend caps what DeepSeek can convert this into, and it means the ranking cannot be read as a commercial result for either side.
- precedent Agents moved while humans stayed put, so migration is now decided workload by workload, and the next release gets judged on whether tool-calling traffic adopts it.
Agentic requests burn about 15 times the tokens of a human request, according to OpenRouter's own data [15], and that multiplier is the mechanism behind everything else. On a blended million in and million out, V4 Flash's cheapest endpoint runs about $0.27 against roughly $35 for GPT-5.5, a gap of about 130x [20]; on output alone it is closer to 167x [19]. The ratio does not change when a workload turns agentic, but the absolute money does, because the same gap is applied to fifteen times the tokens [22]. Anything that runs in a loop and calls tools stops being a model choice and becomes a procurement decision about price per token.
Inside DeepSeek's own traffic the two populations behaved differently, which is the more useful evidence. Human sessions stayed on V3.2 through the first five months of the year [17], so this is not a broad swing in taste; it is agent loops shopping on cost. Hobbyists now route close to a third of their tokens to DeepSeek models [11], and OpenRouter says users at AI-native companies and large organisations also sent considerably more traffic there in early June than they did in January [13].
Two things temper the read. The volume jump did not produce a matching jump in share of spend [6], so the table DeepSeek now tops is the one denominated in the cheapest unit available. And the segmentation carrying the agentic argument is OpenRouter's own: a seven-signal weighted composite scored at the API key level, with tool call rate, turn count and gap timing among the inputs [14]. That is a reasonable proxy rather than an observation of what those keys were doing, and the party publishing it sells routing. The same post puts DeepSeek at nearly 20% at the start of June [4] and at 18% in its January-to-June comparison [3], so treat the figure as a band.
The country arithmetic is the number worth writing down. If American models held about three quarters of tokens through 2025 [7] and Chinese models passed them in early June [8], Chinese share moved from no more than roughly a quarter to above half: a swing of at least 25 percentage points in under eighteen months, on a base that was itself growing [21]. The named Chinese gainers alongside DeepSeek were Xiaomi, Minimax and Tencent, and OpenRouter attributes the offsetting losses to Google and OpenAI [10]. Read plainly, buyers substituted on price at the precise moment their workloads began consuming fifteen times more of the product.
What to watch
- Whether OpenAI or Google answer at the cheap end with a low-cost agentic tier, and how quickly token share responds if they do.
- Whether DeepSeek's human traffic migrates off V3.2 onto V4, which would show the gain is not confined to agent loops.
- Whether the Chinese lead in token share holds past early June or reverses on the next American flagship release.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence52
- Adoption74
- Hype gap+18
- Incentives70
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
DeepSeek began 2026 holding just under 10% of weekly token flow across OpenRouter.
- [2]
As agentic work took hold in February and March 2026, DeepSeek's share fell to 5% of total OpenRouter tokens, squeezed by proprietary models above and other open-source LLMs below.
- [3]
A direct January-to-June 2026 comparison shows DeepSeek's OpenRouter token share going from 9% to 18%.
- [4]
By the start of June, DeepSeek had earned nearly 20% of token share and has been the top model on OpenRouter since mid-May.
- [5]
DeepSeek V4 Flash on the cheapest endpoint costs $0.09 input and $0.18 output per million tokens; GPT-5.5 is priced at $5 input and $30 output per million tokens.
- [6]
The upsurge in V4 token volume has not resulted in an identical spike in share of spend.
- [7]
In 2025, models from the US were responsible for about three quarters of tokens used.
- [8]
Chinese models surpassed American ones in token share as of early June 2026, with only a tiny share going to models from other countries.
- [10]
Chinese open-source model authors including Xiaomi, Minimax and Tencent all saw token share rise over the past six months, which OpenRouter says appears to have come at the expense of Google and OpenAI.
- [11]
Hobbyist users now route nearly a third of all their tokens to DeepSeek models.
- [12]
From its 5% trough to 18% in June, DeepSeek's token share multiplied by about 3.6 times.
- [13]
Users at AI-native companies and at large organizations sent much more DeepSeek token traffic in early June than at the start of the year.
- [14]
OpenRouter segments token traffic at the API key level into Agentic, Mixed and Human using a 7-signal weighted composite score whose inputs include tool call rate, turn count and gap timing.
- [15]
Agentic work burns about 15x more tokens per request than normal human AI usage, according to OpenRouter data.
- [16]
Agentic tokens surpassed human tokens around February 1st 2026, and at first agentic work was concentrated in a few model authors that did not include DeepSeek.
- [17]
Human activity within DeepSeek token usage remained tied to DeepSeek V3.2 through the first five months of the year.
- [18]
By the end of May, a month after release, V4-Flash comprised 70% of agentic token flow for DeepSeek usage.
- [19]
GPT-5.5 output tokens cost about 167 times V4 Flash's cheapest output endpoint.
- [20]
For one million input plus one million output tokens, V4 Flash's cheapest endpoint costs $0.27 versus $35 for GPT-5.5, a gap of about 130x.
- [21]
Chinese-authored models went from no more than about a quarter of tokens in 2025 to more than half by early June 2026, a swing of at least 25 percentage points.
- [22]
Because an agentic request consumes about 15x the tokens of a human request, the absolute per-request price difference between the two endpoints is roughly 15 times larger for agentic work, while the ~130x ratio stays constant.
Sources
1 independent publisher whose own reporting we read for this story.
- DeepSeek V4 Is Earning Agentic Token Share — OpenRouter Blog
openrouter.ai
1 article · August 25, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- Open-Weight Model AdoptionFollow
- US-China model competition and distillation disputesFollow
- Inference Pricing EconomicsFollow
- Model routing marketplacesFollow
- Agentic WorkloadsFollow
- Usage Telemetry MethodologyFollow