September 23, 2026
In Next.js, the choice shifted to price: three models scored 97%
On September 22, GPT-6 Sol, Claude Opus 5.5, and Claude Fable 5.1 each scored 97% in a fresh Next.js evaluation. Grok 4.7 scored 94%, but is 2–7 times cheaper. Next.js ranks Opus first by average cost among the three models at 97%.
Guillermo Rauch
@rauchg
We ran fresh Next.js evaluations. The results: ① Opus 5.5 [𝟿𝟽%] ② GPT-6 Sol [𝟿𝟽%] ③ Fable 5.1 [𝟿𝟽%] ④ Grok 4.7 [𝟿𝟺%] Notably, Grok is 2–7 times cheaper.
GPT-6 Sol debuts at 97% on Next.js evals, tying the highest success rate. Claude Opus 5.5 also reached 97%, joining Claude Fable 5.1. Opus ranks first among the three on average cost. View the leaderboard ↓ https://nextjs.org/evals
· 3.5M views
It is a framework built on React with routing, a server, builds, and deployment. v0, Lovable, Bolt, and Replit often use it by default, while Next.js DevTools MCP gives an agent access to an app's routes, logs, and errors.
A shift in 11 days. On September 11, Claude Fable 5.1 posted 97% at an average price of $0,72 per task. Fable 5 with reasoning set to high scored 77% and cost $1,78. GPT-6 Sol and Opus 5.5 now stand alongside it among the leaders.
How they calculate it. Next.js uses pass@4: a model gets up to four attempts per task and stops after its first success. Timeouts and infrastructure failures are not counted as model errors.
Pricing through Gateway. Grok 4.7 costs $1,20 per million input tokens and $3,60 per million output tokens, with a 40% promotional price in effect through September 27. It has a 500 thousand-token window and four reasoning levels. Claude Opus 5.5 costs from $4 and $20 per million tokens, with a 1 million-token window and a maximum of 128 thousand output tokens. Claude Fable 5.1 costs $10 and $50, while cache reads cost $0,25 per million tokens, 75% cheaper than Fable 5.
Vercel suggests connecting Grok 4.7 through `vercel ai-gateway setup` in Cursor, Codex, Amp, OpenCode, or fx.
The Vercel repository describes rerunning the evaluation with `pnpm install --frozen-lockfile`, `pnpm sync-evals`, `.env.local`, `pnpm preflight`, and `pnpm eval:smoke claude-opus-5`.
Source
