Kimi K2.7 Code vs GPT-5.5 pricing

Kimi K2.7 Code is substantially cheaper per token, but lower price is not proof of equal coding quality. Prices and source records checked .

The short answer

At the tracked hosted rate, Kimi K2.7 Code costs $0.95 input, $0.19 cached input, and $4.00 output per 1M tokens. GPT-5.5 costs $5.00, $0.50, and $30.00. In the three coding-agent workload shapes below, the GPT-5.5 bill is 6.3× to 7.4× the Kimi bill.

API price per 1 million tokens

Model Input Cached input Output Context
Kimi K2.7 Code $0.95 $0.19 $4.00 262,144
GPT-5.5 $5.00 $0.50 $30.00 1.05M

Kimi pricing uses the Novita-hosted canonical route. GPT-5.5 prices are standard short-context rates; OpenAI charges more when input exceeds 272K tokens. Kimi's official model card reports a 256K context window, while OpenAI reports 1.05M for GPT-5.5.

Coding-agent workload cost table

These are deterministic token-cost examples, not measured end-to-end job costs. Real bills also depend on retries, reasoning tokens, tool loops, and cache hit rate.

Workload Kimi K2.7 Code GPT-5.5 Kimi saving
Uncached coding session 1M input + 0.2M output $1.75 $11.00 84%
Cached-prefix session 0.2M fresh + 0.8M cached + 0.2M output $1.14 $7.40 85%
Output-heavy agent run 0.25M input + 1M output $4.24 $31.25 86%

Benchmark caveat: cheaper does not mean GPT-5.5-level

The viral three-prompt canvas physics test is anecdotal, not a reproducible benchmark. Moonshot's own Kimi K2.7 Code model card shows GPT-5.5 ahead on all six published coding and agentic comparisons: Kimi Code Bench v2, Program Bench, MLS-Bench Lite, Kimi Claw 24/7 Bench, MCP Atlas, and MCPMark Verified. Several are vendor-created benchmarks and the models ran through different agent harnesses, so the table is useful evidence—not a universal quality ranking.

Example: Moonshot reports 62.0 for Kimi versus 69.0 for GPT-5.5 on Kimi Code Bench v2, and 81.1 versus 92.9 on MCPMark Verified.

Which model is the better buy?

Choose Kimi K2.7 Code when token budget is the constraint and your own repository-level evaluation says its pass rate is good enough. Choose GPT-5.5 when the expected value of higher task completion outweighs a much larger token bill, or when you need more than 256K context. The correct comparison is cost per successful task, not cost per token alone.

For broader decisions, compare all OpenAI API prices, scan the cheapest AI API ranking, run your workload through the token cost calculator, or use the model comparison calculator.

Sources and freshness

Price cells are read at build time from the daily-maintained canonical pricing dataset. Last canonical update: .