Innovation & Tech

Baidu's Token Exchange Hits 700 Billion Tokens a Day After One Month

Baidu's Token Exchange, launched November 12, 2026, already averages 50 million daily calls and 700 billion tokens. ERNIE 5.5 Flash Preview, agent upgrades, cloud and chips rounded out Baidu World.

By Grace Kim

4 min read

Baidu Launches Token Exchange for Verified AI Model Services
Baidu Launches Token Exchange for Verified AI Model ServicesBru-nO / pixabay

What's News

  • Baidu's Token Exchange averages over 50 million calls and 700 billion tokens per day after one month.
  • ERNIE 5.5 Flash Preview, released November 11, 2026, has 280 billion total parameters and a 1-million-token context window.
  • ERNIE 5.5 Flash Preview pricing: RMB 1 per million input tokens, RMB 4 per million output tokens; Apache 2.0 weights due by November 25, 2026.
  • Baidu's daily active agents rose from 5,000 to 37,000 in a year; weekly completed tasks grew from 84,000 to over 1 million.
  • Kunlunxin's M300 flagship chip is expected in 2027; the M100 inference chip has shipped.

Baidu's new Token Exchange marketplace is already averaging more than 50 million calls and over 700 billion tokens a day, just one month after launch. The company announced the figures at Baidu World 2026 on November 12, 2026, where it also rolled out upgrades spanning models, agents, cloud and chips.

Token Exchange gives businesses and individual users a single place to compare prices and performance across model providers, then buy access. More than 100 model resources have been onboarded so far, Baidu says.

How does the exchange verify what it sells?

Baidu built the marketplace around verification that doesn't stop at listing. The process compares model fingerprints and tests each provider's capability, performance and availability before it goes live.

Oversight continues after that. Providers receive credit ratings based on regular spot checks and real-time, end-to-end call monitoring. The design pairs upfront verification with continuing supervision — a listing is treated as an ongoing commitment, not a one-time check.

That structure matters for the enterprise buyers Baidu is chasing. The open question, as the company itself frames it, is whether verification and credit ratings can turn early usage into a trusted channel for companies choosing among competing model providers.

What's new in ERNIE 5.5 Flash Preview?

One day before Baidu World, on November 11, 2026, Baidu released ERNIE 5.5 Flash Preview. It is the first post-trained model built on the earlier ERNIE 5.5 Flash Base, and the numbers are specific:

  • 280 billion total parameters
  • 14 billion parameters activated per token
  • A 1-million-token context window

Baidu positions the model for coding, 3D modeling, knowledge work and content creation. It says the Preview balances capability with inference efficiency, citing competitive benchmark scores and fast decoding.

The demos focused on work inside an agent harness. In one test, the text-only model spent 11 hours on an Unreal Engine task: it wrote a third-person game in C++ and converted playtest problems — including a hovering character and a misplaced camera — into measurable software tests, because it couldn't see the screen itself.

Pricing is aggressive. The API runs through Qianfan at RMB 1 per million input tokens and RMB 4 per million output tokens, with cached input at RMB 0.04 per million. Baidu plans to release the model weights under Apache 2.0 by November 25, 2026, with a caveat: this intermediate post-training checkpoint may still fall short in some capabilities while training continues.

Two application-specific models shipped alongside it. ERNIE Wizshot Micro-Expression combines text, images and speech for expressive video creation. ERNIE Light Deep Thinking handles complex search requests and tool calls by pairing a fast-thinking model that serves users with a slow-thinking model that continuously refines it. Baidu says its Sparse Attention technology compresses historical memory during decoding, cutting inference costs without compromising performance.

What did Robin Li say about the 'intelligent organization'?

Robin Li, Baidu's co-founder and CEO, framed the whole portfolio around a shift from fixed hierarchies to organizations built around tasks. He laid out three principles:

  • Break down job boundaries
  • Convert tacit knowledge into shared organizational assets
  • Create measurable closed-loop processes that support continuous learning

Baidu's internal numbers make the adoption case. Average daily active agents rose from 5,000 to 37,000 over the past year. Weekly completed tasks jumped from 84,000 to more than 1 million.

Which agents got upgraded?

DuMate is now the unified gateway to Baidu's specialized agents, and its Harness engine and Agent API are open to developers. The DuMate team said it iterated more than 160 times in six months and used the agent to generate 90% of the team's own code.

KOOKO, upgraded from GenFlow 3.0, combines public information, professional databases and users' own content to produce research reports, slide decks and websites.

Famou 3.0 targets production operations by combining an agent platform with forward-deployed engineering services. Baidu reports customers across 25 of China's 41 industrial categories — a coverage rate of roughly 61%.

What's happening under the stack?

The infrastructure announcements connect the silicon to the marketplace. Baidu AI Cloud launched the Tianchi 1024-card Supernode, an all-optical interconnect system built for high-speed data transmission.

Kunlunxin, Baidu's AI chip business, said its M100 inference chip has shipped. The flagship M300 is expected in 2027 with improvements in computing power, memory and interconnection.

Taken together, the announcements link Baidu's silicon and cloud infrastructure to foundation models, workflow agents and a marketplace for model services. Token Exchange is the piece that extends the stack beyond Baidu's own products — and its 700-billion-token daily volume suggests demand exists. The test now is trust: whether credit ratings and continuous monitoring can keep enterprises buying.

Source: unite.ai

Share this article:

More from Grace Kim

Grace Kim

Show full bio

Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

1,985 articles

Related articles

« Previous articleNext article »