Updated June 11, 2026

AI Power Rankings

IndependentΒ·Frontier AI lab index
← Companies
πŸ‡¨πŸ‡³

Baidu

CN

ERNIE 5.0, Ernie Bot, Baidu AI Cloud, Kunlun chips.

Efforts: ERNIE 5.0, Ernie Bot, Baidu AI Cloud, Kunlun chips
Rank
#12
Structural
7.63
Unweighted
7.57
vs #1
-1.69

Profile across ten areas

Area breakdown

🧠Model Quality
Core Engine Β· 18%
ERNIE 5.1 ranked #13 globally and #1 among Chinese models on LMArena (1,476), is #4 globally on the search leaderboard, tops the Legal AI board, and the omni-modal 2.4T-parameter ERNIE 5.0 base covers text/image/audio/video. Counterpoint: Arena Elo flatters conversational quality β€” on agentic coding benchmarks like SWE-bench Pro, ERNIE is absent from the top ranks where Kimi, GLM, and Qwen compete.
β€” Claude Fable 5
Claude: 7.7Gemini: 8.8GPT-5.5: 7.5Grok: 6.8
7.7
panel mean
πŸ”¬Research
Core Engine Β· 12%
The 'Once-For-All' elastic pre-training claim (ERNIE 5.1 at 6% of peer training cost), a decoupled asynchronous RL pipeline, and the self-evolving Famou Agent 2.0 (10.21% lift at an automated port) represent substantive efficiency research. Counterpoint: the 94% cost-reduction claim lacks independent verification, and Baidu's open-source contributions and global research influence trail Qwen and DeepSeek.
β€” Claude Fable 5
Claude: 7.1Gemini: 8.4GPT-5.5: 7.7Grok: 6.9
7.5
panel mean
πŸ—οΈProduct
Delivery Β· 10%
Ernie Bot at 200M MAU, AI woven directly into China's dominant search engine, and a broad Create 2026 agent portfolio (DuMate, Miaoda β€” 90% self-generated code β€” Yijing, Famou 2.0) on a rebranded full-stack agent cloud. Counterpoint: the agent products are weeks old with unproven adoption, and Baidu's developer/API mindshare outside China is minimal.
β€” Claude Fable 5
Claude: 7.3Gemini: 8.4GPT-5.5: 7.8Grok: 7.3
7.7
panel mean
πŸ’°Business
Accelerants & Stabilizers Β· 5%
Q1 2026 beat across the board: AI-powered Core revenue +49% YoY to RMB13.6B (52% of General Business for the first time) and AI Cloud +79% YoY to RMB8.8B β€” the clearest proven AI monetization among Chinese pure-plays. Counterpoint: legacy ad revenue fell 29% YoY, so AI growth is partly racing a melting core, and the June 9 US CMC-list designation adds geopolitical overhang.
β€” Claude Fable 5
Claude: 7.2Gemini: 9.0GPT-5.5: 7.8Grok: 7.5
7.9
panel mean
⚑Compute
Core Engine Β· 14%
Kunlunxin custom silicon (M100 in 2026, M300 in 2027) already trained ERNIE 5.1 in production, and the Kunlunxin-backed AI Cloud is monetizing at RMB8.8B/quarter growing 79% β€” proof the full stack works at scale. Counterpoint: Kunlunxin volumes and performance versus Nvidia remain unverified, the CMC listing could complicate supply chains, and capacity trails Alibaba's hyperscale footprint.
β€” Claude Fable 5
Claude: 7.5Gemini: 9.0GPT-5.5: 7.8Grok: 7.0
7.8
panel mean
πŸ›‘οΈSafety
Accelerants & Stabilizers Β· 6%
As a NASDAQ/HKEX-listed company Baidu carries real disclosure and governance obligations, has long operated under China's content-safety regime, and held an orderly June 5 AGM amid the transition. Counterpoint: the June 9 addition to the US DoD Chinese Military Companies list (which Baidu disputes) raises trust friction, and published alignment/interpretability research is thin by frontier-lab standards.
β€” Claude Fable 5
Claude: 4.2Gemini: 7.7GPT-5.5: 6.9Grok: 6.4
6.3
panel mean
πŸ‘₯Talent
Delivery Β· 10%
Robin Li's early and consistent AI-first conviction, a deep veteran engineering bench, and a coherent strategic reframe around 'Daily Active Agents' show directional clarity. Counterpoint: Baidu has been a chronic talent exporter to startups (including ex-Baidu founders across China's AI scene), and a 20-year-old company moves slower than the labs it competes with.
β€” Claude Fable 5
Claude: 6.4Gemini: 8.1GPT-5.5: 7.6Grok: 7.1
7.3
panel mean
🌐Distribution
Delivery Β· 8%
Search dominance in China plus 200M Ernie Bot MAU and growing enterprise cloud anchoring give Baidu mass default distribution domestically. Counterpoint: international reach is negligible, ERNIE's open-source/community presence is far weaker than Qwen's 942M downloads or Kimi's OpenRouter rank, and the CMC listing further chills overseas adoption.
β€” Claude Fable 5
Claude: 6.8Gemini: 8.9GPT-5.5: 8.5Grok: 7.6
8.0
panel mean
🏰Data & Moats
Core Engine Β· 12%
Decades of Chinese search queries, maps, and now agent-era enterprise cloud workloads form a proprietary data flywheel, reinforced by regulatory licenses and the 'enterprises can't build their own AI infra' cloud lock-in thesis (May 19). Counterpoint: search data value erodes as AI answers cannibalize the ad model (legacy -29% YoY), and the moat is entirely domestic.
β€” Claude Fable 5
Claude: 7.0Gemini: 9.0GPT-5.5: 8.3Grok: 7.8
8.0
panel mean
πŸš€Momentum
Accelerants & Stabilizers Β· 5%
ERNIE 5.1 (April preview, May launch), Create 2026's agent portfolio, and a beat quarter with AI crossing 50% of core business mark genuine acceleration. Counterpoint: developer narrative energy in China has shifted toward Qwen, Kimi, and DeepSeek, and the CMC designation plus shrinking legacy revenue temper the trajectory.
β€” Claude Fable 5
Claude: 7.1Gemini: 8.5GPT-5.5: 7.7Grok: 6.7
7.5
panel mean

News log

Baidu β€” News Log

Tracked for AI Power Rankings scoring. Covers model releases, benchmarks, pricing, funding, partnerships, infrastructure, and policy changes.


2026-06-09 β€” Baidu added to U.S. Department of Defense's Chinese Military Companies (CMC) List; disputes inclusion

Published: 2026-06-09 | Logged: 2026-06-11T00:00Z | Area: Safety & Alignment, Business & Market

On June 9, 2026, Baidu disclosed it had been added to the U.S. Department of Defense's list of Chinese Military Companies (the "CMC List"). Baidu stated it is neither a Chinese military company nor a military-civil fusion contributor and said there is no justification for the inclusion. The company emphasized that the list is not a sanctions list, does not restrict trading in its securities on NASDAQ or the Hong Kong Stock Exchange, and that the associated U.S. government procurement limitations will not impact its business.

Scoring impact: Negative for Safety & Alignment / geopolitical-risk dimension of Business & Market (heightened U.S.-China regulatory overhang and reputational friction), though Baidu asserts limited near-term operational impact. Adds to the policy-risk backdrop around its AI ambitions and Kunlunxin chip program.

Sources:



2026-06-05 β€” Baidu holds 2026 Annual General Meeting

Published: 2026-06-05 | Logged: 2026-06-11T00:00Z | Area: Business & Market, Talent & Org

Baidu held its Annual General Meeting of shareholders on June 5, 2026, the routine governance milestone at which shareholders vote on board composition and corporate matters. The meeting follows a quarter in which AI-powered Core revenue first crossed 50% of Baidu's General Business and management reframed strategy around "Daily Active Agents."

Scoring impact: Neutral-to-mildly-positive governance signal for Business & Market and Talent & Org (orderly board/governance cadence amid the AI-led business transition); not a product or model catalyst.

Sources:



2026-05-19 β€” Baidu frames cloud as the way to monetize AI infra most enterprises can't build

Published: 2026-05-19 | Logged: 2026-06-11T00:00Z | Area: Business & Market, Compute & Infra, Distribution & Reach

Following Q1 results, Baidu executives argued that most enterprises cannot build their own large-scale AI infrastructure, positioning Baidu AI Cloud's Kunlunxin-backed, full-stack agent platform as the natural place for that demand to land. Coverage characterized this as Baidu "saying the quiet part out loud" β€” that the inability of customers to self-build is precisely what makes cloud the monetization layer for its AI stack. The framing reinforces the Q1 AI Cloud line (RMB8.8B, +79% YoY).

Scoring impact: Positive for Business & Market and Distribution & Reach (clearer monetization thesis for enterprise AI demand). Compute & Infra reinforced (Kunlunxin-backed stack as the differentiator).

Sources:



2026-05-18 β€” Baidu Q1 2026 results: AI-powered Core +49% YoY; AI Cloud +79% YoY; AI is now 52% of general business

Published: 2026-05-18 | Logged: 2026-05-19T09:00Z | Area: Business & Market, Compute & Infra, Distribution & Reach

Baidu reported Q1 2026 revenue of RMB32.1B, beating estimates on revenue, adjusted profit, and operating profit. Core AI-powered Business revenue grew 49% YoY to RMB13.6B, exceeding 50% of Baidu General Business revenue for the first time (52%). AI Cloud Infra revenue hit RMB8.8B, up 79% YoY and 52% sequentially β€” the standout line and the clearest validation yet that enterprise demand is monetizing Baidu's Kunlunxin-backed stack. Legacy Business (largely ad-driven) declined 29% YoY to RMB10.2B, underscoring the AI-versus-legacy transition. CEO Robin Li framed the quarter around the May 13 "Daily Active Agents" pivot and ERNIE 5.1's low pre-training cost.

Scoring impact: Strong Business & Market positive (AI line crosses the 50% threshold; AI Cloud growth accelerating off a meaningful base). Compute & Infra signal (Kunlunxin-based AI Cloud monetizing at scale). Distribution & Reach lift as enterprise customers anchor more workloads on Baidu's stack. Partially offset by sharp legacy decline.

Sources:



2026-05-13 β€” Baidu hosts Create 2026 developer conference; expected ERNIE 5.1 technical disclosures and commercialization plans

Published: 2026-05-13 | Logged: 2026-05-13T09:00Z | Area: Product & Platform, Research & Innovation, Business & Market

Baidu opened its annual Create 2026 developer conference May 13–14 in Beijing. The agenda foregrounds industrial applications of the recently released ERNIE 5.1 model, which Baidu claims achieved its pre-training at just 6% of the cost of peer models thanks to a new "Once-For-All" multi-configuration training framework. Founder Robin Li is expected to disclose more technical details on the architecture and lay out commercialization plans, against a backdrop where ERNIE 5.1 currently ranks first among Chinese models on Arena and fourth globally β€” closing in on Gemini 3.1 Pro on creative writing and reasoning, and ahead of DeepSeek-V4-Pro on agent capabilities.

Scoring impact: Potential boost across Product & Platform (industrial deployments, developer tooling), Research & Innovation (training-efficiency disclosures), and Business & Market (commercialization roadmap). The previously logged ERNIE 5.1 launch (2026-05-09) already lifted Model Quality; this event compounds the narrative if technical details validate the 94% cost-reduction claim.

Sources:



2026-05-13 β€” Baidu Create 2026: new agent portfolio (DuMate, Miaoda, Yijing, Famou 2.0); CEO declares "agent era"

Published: 2026-05-13 | Logged: 2026-05-14T09:00Z | Area: Product & Platform, Research & Innovation, Business & Market

At Baidu Create 2026, Baidu unveiled a new generation of agent products: DuMate (general-purpose agent with PC/mobile sync, screen reading, file processing, and business-system orchestration); Miaoda (coding agent with app and enterprise editions β€” 90% of the Miaoda app is itself Miaoda-generated code); upgraded Baidu Yijing digital human platform; and Famou Agent 2.0, a self-evolving agent reportedly delivering a 10.21% performance lift at a major automated port. CEO Robin Li proposed "Daily Active Agents (DAA)" as a new industry metric akin to mobile DAU and said AI competition is moving from model races to agents and "super individuals." Baidu AI Cloud also rebranded as a full-stack AI cloud purpose-built for large-scale agent applications. ERNIE 5.1 training was supported on Baidu's Kunlunxin chip cluster, with ERNIE 5.1 ranked #1 among Chinese models on LMArena text/search.

Scoring impact: Strong Product & Platform expansion (broad agent suite). Research & Innovation positive (self-evolving agent claims). Compute & Infra signal (Kunlunxin chip used in ERNIE 5.1 training). Business & Market positioning sharpens around agents.

Sources:



2026-05-09 β€” Baidu officially releases ERNIE 5.1 at 6% of peers' pre-training cost

Area: Model Quality, Research & Innovation, Compute & Infra

Baidu officially released ERNIE 5.1, claiming pre-training cost at just 6% of similarly-scaled industry models. The model was distilled from ERNIE 5.0 using multi-dimensional elastic pre-training, compressing total parameters to one-third and active parameters to about one-half of the prior generation. Baidu says ERNIE 5.1 tops China search-capability rankings and surpasses DeepSeek-V4-Pro on agent capabilities, with creative writing and reasoning approaching Gemini 3.1 Pro. On LMArena it ranks #4 globally on the search leaderboard (1223 score) and is the only Chinese model in the top tier there. A new decoupled, fully asynchronous RL training pipeline is credited for stability gains.

Scoring impact: Positive for Model Quality (closing the gap to frontier on key tasks), Research & Innovation (training-efficiency technique), and Compute & Infra (lower training capex per quality unit).



2026-04-29 β€” ERNIE 5.1 Preview launches as #1 Chinese model on LMArena

Published: 2026-04-29 | Logged: 2026-05-02T09:00Z | Area: Model Quality, Business & Market

Baidu released ERNIE 5.1 Preview, which shot to #13 globally on LMArena's Text Arena (score 1,476), becoming #1 Chinese AI model on the leaderboard. Beat DeepSeek-V4-Pro which launched just days earlier. ERNIE 5.1 Preview is a lighter, cheaper version atop the 5.0 foundation. Also topped the Legal AI leaderboard worldwide. Full launch confirmed for Baidu Create 2026 developer conference.

Scoring impact: Strong Model Quality boost (#1 Chinese model, global top 15). Positive for Business & Market (competitive positioning, efficiency gains).

Sources:



2026-01-15 β€” Baidu's ERNIE 5.0 ranks first among Chinese models; custom AI chips announced

Published: 2026-01-15 | Logged: 2026-05-02T09:00Z | Area: Model Quality, Compute & Infra, Product & Platform

Baidu's ERNIE 5.0, a 2.4 trillion parameter omni-modal model (text, images, audio, video), ranked first among Chinese models and eighth globally on LMArena text benchmarks, ahead of GPT-5.1-High and Gemini-2.5-Pro. The model uses an ultra-large MoE architecture with less than 3% of parameters active per inference. Baidu also announced two custom AI chips: Kunlunxin M100 (2026) and M300 (2027). The Ernie Bot assistant has reached 200 million monthly active users, and Baidu's 2026 strategy integrates AI directly into search rather than as a separate product.

Scoring impact: Positive for Model Quality (top Chinese model ranking). Compute & Infra boosted by custom chip development. Product & Platform strengthened by 200M MAU and search integration strategy.

Sources:



← Higher rank
πŸ‡ΊπŸ‡Έ Apple
#11 Β· 7.65
Lower rank β†’
πŸ‡«πŸ‡· Mistral AI
#13 Β· 7.36