OpenRouter Rankings July 2026:
Who's Actually Winning the AI Model Race

OpenRouter ranks by token volume — essentially what developers pay to run in production, not who wins a leaderboard. That gap between usage and capability is the frame you need before reading July's board. Compared with our June 2026 recap, three mistakes show up constantly:

• Treating the daily volume #1 as "best model": A cheap, fast model wired into one high-traffic consumer app can outrank a more capable model reserved for the hardest 10% of work. On July 25, Xiaomi's Mimo V2.5 led at roughly 1.4 trillion tokens/day — a price-and-reach story, not a universal quality crown.
• Ignoring day-to-day volatility: Rankings move fast. Claude Opus 4.8 sat in the top 10 on July 24 weekly data; by July 25 Ling 3.0 Flash had pushed it out of the top 12. Monthly series beat single-day snapshots.
• Looking only at models, not apps: Hermes Agent alone holds about 45% of tracked app token share. Roleplay and companion apps move serious volume that enterprise AI coverage rarely mentions — a hidden market that distorts "who is using what."

Bottom line: the rankings answer who pays the everyday token bill; spend-by-task-category data answers who earns premium pricing on the hardest work. Read both.

Official boards update daily — verify current figures before citing:

https://openrouter.ai/rankings

https://openrouter.ai/apps

As of July 25, 2026, the top three models by daily token volume are Xiaomi Mimo V2.5 (1.4T/day), DeepSeek V4 Flash (943.9B/day), and Tencent Hy3 (590B/day). Seven of the top ten spots belong to Chinese labs; Nemotron 3 Ultra (NVIDIA), Claude Sonnet 5, and Gemini 3 Flash Preview still hold ground for the US side.

At provider level, Chinese-origin labs (DeepSeek, Xiaomi, Tencent, Z.ai, MiniMax, Moonshot, Alibaba) now account for roughly 46% of identified token volume — up from under 2% a year ago. US-origin models (OpenAI, Anthropic, Google combined) sit near 30–36%, down from about 70% a year ago.

Pricing drives the shift: DeepSeek V4 Flash lists at roughly $0.05–0.14/M input tokens versus GPT-5.5 near $5.00/M — a gap of about 35×. DeepSeek remains the most stable #1 provider by share (16–18%), while the "model of the month" crown keeps rotating — MiniMax M2.5 in February, MiMo-V2-Pro in April, Mimo V2.5 in July.

OpenRouter's spend-by-task-category breakdown tells a different story than raw token count: general chat 35.7%, agentic workflows 30.4%, code 26.5%, data work 7.5%. Drill into the hardest bucket — classification and complex reasoning — and Claude Sonnet 4.6 and Claude Opus 4.7 tie at 13.5% of spend each, with GPT-5.5 third at 11.6%. The cheap open models dominating volume charts barely register here.

The market is bifurcating: inexpensive Chinese open-weight models absorb high-volume, error-tolerant workloads (chat, creative writing, roleplay, routine coding assist), while closed frontier models still command pricing power on hard, low-error-tolerance work. Anthropic's July 24 launch of Claude Opus 5 topped FrontierBench v0.1 at 43.3% (versus GPT-5.6 Sol's 37.5%) while holding Opus-tier pricing at $5/$25 per million tokens.

Model rankings show which "brain" is popular. The app leaderboard (openrouter.ai/apps) shows what that brain is doing in production.

• Cline → Roo Code → Kilo Code are three generations of the same open-source lineage; the youngest fork, Kilo Code, has now overtaken both ancestors — first-mover advantage in dev tooling does not last.
• Roleplay and companion apps (Janitor AI, ISEKAI ZERO, SillyTavern, HammerAI) collectively move serious volume. The OpenRouter × a16z State of AI report found creative roleplay accounts for more than half of all open-model usage — if your view of AI usage comes only from enterprise headlines, you are missing half the market.

Based on July's trajectory and surrounding industry context, here is what we expect heading into August:

1. Chinese open-weight combined share likely keeps climbing toward or past 50% unless a major US provider makes a real pricing move.
2. The "model of the month" title keeps rotating — Xiaomi, DeepSeek, Tencent, Z.ai, MiniMax, and Moonshot ship and re-price fast enough that a new #1 by August would not be surprising.
3. Anthropic may ship a cheaper, volume-focused tier rather than relying on Opus 5 alone — Opus 5 is Anthropic's fourth flagship in under two months (after Mythos 5, Fable 5, and Sonnet 5).
4. Kimi K3's 1.4-terabyte open weights will likely see community quantization within 2–4 weeks, following prior mega-release patterns.
5. Security and governance become real selection criteria: OpenAI's unreleased-model sandbox escape, the proposed US "AI Kill Switch Act," and expected White House pre-release review frameworks will push vendor safety track records into formal enterprise scorecards.

Independent developers and small teams: OpenRouter remains the fastest way to A/B test dozens of models behind one API key. Start coding workloads on DeepSeek V4 Flash and GLM 5.2; reserve premium closed models for steps where cheaper models actually fail. Test your own latency — OpenRouter is not optimized for every region or compliance need.

Infrastructure leads: Do not select models by usage rank alone — it reflects price sensitivity, not fitness for your workload. Route high-volume/low-risk work to cheap models and hard/high-stakes work to frontier closed models; add vendor safety history to your scorecard.

Agent and coding tool builders: The Cline → Kilo Code fork chain is a reminder that moats in open-source dev tooling are thinner than they look. If your product touches entertainment or companion use cases, do not underestimate that segment's volume.

1. Register on OpenRouter and build a cost dashboard: Create an API key, tag spend by model and project, and review the Rankings page weekly so your default model does not fossilize.
2. Define tiered routing rules: Split workflows into Tier-0 (autocomplete/summary), Tier-1 (multi-file refactors), Tier-2 (long-horizon agents). Point Tier-0 at DeepSeek V4 Flash or Mimo V2.5; Tier-2 at Claude Opus 5.
3. Enable swappable model config in CLI tools: Claude Code, Kilo Code, Hermes Agent, and OpenClaw accept OpenRouter model IDs via env vars or config files — never hard-code a single endpoint.
4. Reserve a self-host path for open weights: For compliance-sensitive data, evaluate on-prem MiniMax M3, DeepSeek V4, GLM 5.2, or Kimi K3; on Mac, Ollama or LM Studio works as an offline fallback.
5. Run a 24/7 orchestration layer on Mac: Agent gateways, scheduled batch jobs, and SSH tunnels should not depend on a laptop lid. Use launchd or bare-metal Mac so routing and log collection stay online.
6. Lock the August release calendar and prep A/B cutovers: Watch for Anthropic's volume tier, Kimi K3 community quant builds, and August ship cadence from Chinese labs; stage shadow traffic and rollback switches before launch week.

• Xiaomi Mimo V2.5 daily tokens: about 1.4T, model rank #1 on July 25
• Chinese vendor combined share: about 46% (under 2% a year ago); US big three near 30–36%
• DeepSeek V4 Flash vs GPT-5.5 input gap: about 35× ($0.05–0.14/M vs ~$5/M)
• Hermes Agent app share: about 45%, largest single app on the platform
• Claude Opus 5 FrontierBench v0.1: 43.3%, above GPT-5.6 Sol's 37.5%

The story to remember from July: capability and popularity are diverging. Chinese open-weight models bought half the market with price. US closed-frontier labs are defending the other half with pricing power on hard tasks and safety credibility. The more useful question for builders is not "who is #1 this week," but which side of the barbell your workload actually belongs on.

Deploying hybrid routing on Mac surfaces familiar gaps: laptop sleep drops agents and OpenRouter callbacks; a Linux VPS cannot run Claude Code's macOS-native sandbox; virtualized Mac instances pay Metal and file-permission penalties versus bare metal; buying a maxed Mac Studio adds procurement lead time and depreciation. Teams needing 24/7 uptime with elastic M4/M4 Pro sizing for agent orchestration and iOS CI/CD should look at CALMVPS bare-metal Mac rental — dedicated Apple Silicon, roughly 120-second provisioning, and day/week/month/quarter billing — so you trust OpenRouter boards for model choice while keeping the routing layer on macOS that never sleeps. See pricing for hardware tiers and help center for remote access setup.