LongCat-2.0 vs Claude Code

API pricing, benchmarks, and how to run LongCat-2.0 inside Claude Code.

Production Replacement Eval Official Setup Docs

API Pricing vs Claude Sonnet 5

Model / Tier Input ($/M) Output ($/M) Notes
LongCat-2.0 (standard) $0.75 $2.95 Free cached-context reads
LongCat-2.0 (promo) $0.30 $1.20 Launch promotion
Claude Sonnet 5 (intro pricing) $2.00 $10.00 Introductory API tier

On output tokens, LongCat-2.0 standard pricing is roughly 3.4× cheaper than Claude Sonnet 5 intro rates ($2.95 vs $10/M). Against GPT-5.5's $30/M output, LongCat's gap is closer to 10× — see our GPT-5.5 comparison for that breakdown.

Benchmark comparison

Benchmark LongCat-2.0 Claude / GPT reference Interpretation
SWE-bench Pro 59.5 GPT-5.5: 58.6 Slightly ahead of GPT-5.5 on reported scores; run your own eval before switching from Claude
FORTE 73.2 Claude Opus 4.6: 73.2 (tie); GPT-5.5: 77.8 Matches Opus 4.6 on workflow sim; does not beat GPT-5.5 here
Terminal-Bench 2.1 70.8 — Strong terminal signal; compare on your CLI tasks directly

LongCat-2.0 leads on coding benchmarks (SWE-bench Pro, Terminal-Bench) in public reports but trails GPT-5.5 on FORTE (general workflow simulation). Review both before moving a Claude Code pipeline to LongCat.

Official Claude Code Integration

LongCat publishes first-party docs for running LongCat-2.0 through Claude Code's Anthropic-compatible API layer:

  • Edit ~/.claude/settings.json
  • Set ANTHROPIC_BASE_URL to https://api.longcat.chat/anthropic
  • Set ANTHROPIC_MODEL (and related defaults) to LongCat-2.0
  • Provide your LongCat API key as ANTHROPIC_AUTH_TOKEN

Full step-by-step instructions, env keys, and test commands: longcat.chat/platform/docs/ClaudeCode.html

The same platform also documents integration for Kilo Code, OpenCode, OpenClaw, and Codex — see the LongCat API docs hub for each tool's page.

For production fit — cost, failure modes, and benchmarks — see the Claude Code production evaluation and agent workflow decision guide.

Related Comparisons