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

Why gut-feel model picks fail in July 2026

The leaderboard today looks nothing like a year ago. These friction points are forcing teams to rebuild how they select models:

• Daily rank volatility: The same model can move several spots overnight (Mimo V2.5's Top 10 order shifted between July 24 and 25). A single snapshot is not a strategy.
• Usage is not quality: A cheap model wired into one high-traffic consumer app can outrank a more capable model reserved for the hardest 10% of work.
• Invisible demand: Roleplay and companion apps move serious open-model volume that enterprise AI coverage almost never mentions.
• Pricing gap widened: DeepSeek V4 Flash lists at roughly $0.05–$0.14/M input tokens; GPT-5.5 sits around $5/M — roughly a 35× spread.
• Security entered the scorecard: OpenAI's sandbox-escape disclosure this week, plus proposed US "AI Kill Switch" legislation, pushes vendor safety track records into formal procurement reviews.

OpenRouter model token volume Top 12 (July 25)

Seven of the top ten daily-volume models on July 25 come from Chinese labs. DeepSeek remains the most stable #1 provider by share (16–18%), but the "model of the month" crown keeps rotating.

Provider share and pricing: China ~46%, US ~30–36%

At the provider level, Chinese-origin labs now account for roughly 46% of identified token volume — up from under 2% a year ago. US-origin models (OpenAI, Anthropic, Google combined) fell from ~70% in mid-2025 to roughly 30–36%. This is pricing math, not geopolitics.

Price comparison explains the traffic migration:

Usage rank is not a quality signal: the barbell market

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 category — classification/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: Chinese open-weight models absorb high-volume, error-tolerant workloads (chat, creative writing, roleplay, routine coding), while closed frontier models defend pricing power on hard, low-error-tolerance work. Claude Opus 5 (July 24) topped FrontierBench v0.1 at 43.3% versus GPT-5.6 Sol's 37.5%, holding Opus-tier pricing at $5/$25 per million tokens — a "we cost more, and we're worth it" bet.

App leaderboard: coding agents rule; roleplay is half the market

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

• Hermes Agent (Nous Research) holds ~45% of tracked app token share — the single largest app on the platform.
• Coding agents fill most of the rest: Kilo Code (~13%), OpenClaw (~9%), Claude Code (~6%), Cline (~1.7%).
• Cline → Roo Code → Kilo Code are three generations of the same open-source lineage; the youngest fork now leads the family in volume. First-mover advantage does not last in dev tooling.
• Roleplay/companion apps (Janitor AI, ISEKAI ZERO, SillyTavern, HammerAI) move serious volume. OpenRouter × a16z's State of AI report found creative roleplay accounts for more than half of all open-model usage.

August outlook: five signals to watch

1. Chinese open-weight combined share likely keeps climbing toward 50% unless a major US provider makes a real pricing move.
2. The "model of the month" title keeps rotating across Xiaomi, DeepSeek, Tencent, Z.ai, MiniMax, and Moonshot.
3. Anthropic may ship a cheaper, volume-focused tier — Opus 5 is already their fourth flagship in under two months.
4. Kimi K3's 1.4TB open weights will likely see community quantization within 2–4 weeks before smaller teams can run them practically.
5. Security and governance become formal selection criteria — sandbox escapes, pre-release review frameworks, and vendor safety track records enter enterprise scorecards.

Six steps to build layered model routing

Do not select models by usage rank alone. These six steps turn OpenRouter data into engineering policy (new to the platform? Start with our OpenRouter API guide):

1. Stamp the data date and set a monthly review cadence: Open openrouter.ai/rankings and record the check date; ranks shift daily — treat this as a recurring series, not a one-off post.
2. Split eval sets by task type: Bucket workloads into chat/creative, agentic workflows, code, and complex reasoning; track acceptance rate, error rate, and P95 latency per bucket.
3. Route volume work to cheap open models: Start with DeepSeek V4 Flash for cost-efficiency and GLM 5.2 for the closest open-weight match to Opus-style planning on tolerant tasks.
4. Reserve closed frontier models for hard steps: Classification, complex reasoning, and high-stakes agent decisions go to Claude Opus 5 / GPT-5.6 — only where cheaper models actually fail.
5. A/B coding agents on the same repo: Compare Kilo Code, Cline, and OpenClaw default backends and fallback chains; log token cost and fix success rate — moats in open dev tooling are thinner than they look.
6. Add vendor safety to your scorecard: Track sandbox incidents, data retention, and permission boundaries; autonomous agent flows need least-privilege defaults and human approval gates.

Hard numbers and primary sources

• ~46% Chinese provider token share: Up from under 2% a year ago — one of the steepest share migrations in AI over the past 12 months, cross-validated against OpenRouter official data and third-party 7-day trackers.
• ~35× input price gap (DeepSeek V4 Flash vs GPT-5.5): ~$0.05–0.14/M versus ~$5/M — the core driver of volume migration to Chinese open models.
• Hermes Agent ~45% app-layer share: Far ahead of #2 Kilo Code (~13%) — personal/CLI agents are the largest single traffic entry point on OpenRouter today.
• Claude Opus 5 FrontierBench v0.1: 43.3%: Ahead of GPT-5.6 Sol at 37.5% at unchanged Opus-tier $5/$25 pricing — per Anthropic's July 24 launch.

Rankings shift daily. Verify current figures before citing.

Primary source — OpenRouter model rankings:

OpenRouter Model Rankings

Primary source — OpenRouter app leaderboard:

OpenRouter Apps Leaderboard

OpenRouter × a16z State of AI report:

OpenRouter State of AI 2025

Anthropic Claude Opus 5 launch:

Introducing Claude Opus 5 — Anthropic

OpenRouter rankings FAQ

Running Kilo Code, Cline, or Hermes Agent comparisons on your daily driver pollutes global config and agent state. A KVMFLUX cloud Mac mini gives you root access, SSH + VNC, and dedicated physical M4 hardware on a daily rental — test DeepSeek V4 Flash versus Claude Opus 5 fallback chains in a clean environment, then walk away. If you're choosing a rental period, daily billing fits these monthly leaderboard-driven validation sprints best.