Best Model for Kilo Code

Claude Sonnet 5.5 is the best main model for Kilo Code: it scores 61.8% on Terminal-Bench 4.0 and costs about $5.33 per 100 coding-agent requests on our measured traffic. Kilo lets you set separate models for the main agent, subagents and small tasks such as commit messages, so Claude Haiku 5.5 ($1.31 per 100 requests) fits the small and subagent slots while Claude Opus 5.5 (64.9%) takes the hardest sessions. Kilo can also sign in with a ChatGPT Plus or Pro plan instead of an API key, which makes GPT-6.1 Sol (58.2%) the natural pick for ChatGPT subscribers.

Read the full Kilo Code review → · updated 2026-10-17

Sources: Kilo model selection docs · Terminal-Bench 4.0 leaderboard · Scale SWE-Bench Pro V2 · Anthropic on Haiku 5.5 · DeepSeek V4.1 Flash release · Cost per request methodology

#1Claude Sonnet 5.5Anthropic · $4.40/1M · 1M ctx

Best overall. The main model for code, plan and debug work: 61.8% on Terminal-Bench 4.0 and 95.8 on Scale's SWE-Bench Pro V2, with a 1M-token context.

Overall 96/100 · Value 59Full breakdown →

#2Claude Opus 5.5Anthropic · $8.80/1M · 1M ctx

Best for long agent runs. #1 on Terminal-Bench 4.0 at 64.9%, for large refactors and long autonomous sessions. Every rate is double Sonnet 5.5's.

Overall 98/100 · Value 51Full breakdown →

#3Claude Haiku 5.5Anthropic · $0.22/1M · 1M ctx

Best value. Anthropic reports 39.2% on Terminal-Bench 4.0 at $0.10/$0.50 per million tokens. A good small and subagent model while its prompts stay under 100K tokens, where the price steps up five times.

Overall 82/100 · Value 97Full breakdown →

#4GPT-6.1 SolOpenAI · $4.40/1M · 1.1M ctx

OpenAI pick. 58.2% on Terminal-Bench 4.0 and 96.1 on Scale's SWE-Bench Pro V2, and usable through a ChatGPT Plus or Pro sign-in in Kilo.

Overall 92/100 · Value 58Full breakdown →

#5DeepSeek V4.1 FlashDeepSeek · $0.22/1M · 1.0M ctx

Budget pick. Open weights at $0.30/$1.20 per million tokens at peak hours and half off-peak. DeepSeek reports 42.3% on Terminal-Bench 4.0 for its October 17 update, up from 31.2%, so it suits routine edits more than hard tasks.

Overall 77/100 · Value 93Full breakdown →

#6GLM 5.3Z.AI · $1.50/1M · 1.0M ctx

Open-weight pick. The top open-weight model on Terminal-Bench 4.0 (41.8%) and 95.6 on SWE-Bench Pro V2. Its cache reads cost more than Anthropic's, so cache-heavy sessions cost about as much as Sonnet 5.5.

Overall 83/100 · Value 69Full breakdown →

What 100 Kilo Code requests cost on each pick

A typical request sends 99,534 input tokens, 87% of them read from cache, and a median session runs 16 requests. Measured on pooled coding-agent requests through our gateway (79,301 requests from 75 users; Kilo Code alone has fewer than 50 users, so its traffic is pooled), Sep 13 to Oct 12, 2026, priced at each vendor's list rates.

Model	100 requests	Median session	Month of regular use
Claude Sonnet 5.5	$5.33	$0.74	$77.42
Claude Opus 5.5	$10.66	$1.47	$155
Claude Haiku 5.5	$1.31	$0.096	$17.73
GPT-6.1 Sol	$6.74	$0.74	$90.04
DeepSeek V4.1 Flash	$0.67	$0.091	$9.13
GLM 5.3	$5.72	$0.71	$92.16

Month of regular use = the median user active on 5 or more days (1,478 requests in 30 days). Methodology · All models ranked by cost per request

Best model for Kilo Code under $19, $49 or $99 a month

The highest-rated model whose list-price cost for a month of regular use (1,478 requests) stays under each budget. Heavier months cost proportionally more.

Monthly budget	Best model that fits	Est. month
Under $19	Claude Haiku 5.5	$17.73
Under $49	Claude Haiku 5.5	$17.73
Under $99	Claude Sonnet 5.5	$77.42

"Highest-rated" uses our editorial quality and agentic scores; candidates are current models with a checked vendor price.

FAQ

What is the best model for Kilo Code?

Claude Sonnet 5.5 as the main model (61.8% on Terminal-Bench 4.0), with Claude Haiku 5.5 for subagents and small tasks and Claude Opus 5.5 for the hardest sessions. Kilo also keeps a live board at kilo.ai/models of the models its users are succeeding with.

Can Kilo Code use different models for different tasks?

Yes. Kilo has separate settings for the main model, a small model for titles, commit messages and prompt enhancement, a subagent model, and models for autocomplete, memory and compaction.

Does Kilo Code support open-weight models?

Yes. Kilo supports more than 30 providers, including OpenRouter, DeepSeek and Ollama, so GLM 5.3 and DeepSeek V4.1 Flash run hosted or self-hosted.

Is GLM 4.6 still the best value in Kilo Code?

GLM 5.3 has replaced it as Z.AI's current model and holds the top open-weight score on Terminal-Bench 4.0 (41.8%). GLM 4.6's 202K context is also smaller than about a fifth of the coding-agent requests we measured.

How much do 100 Kilo Code requests cost?

At vendor list prices, 100 average coding-agent requests cost: Claude Sonnet 5.5 $5.33, Claude Opus 5.5 $10.66, Claude Haiku 5.5 $1.31, GPT-6.1 Sol $6.74, DeepSeek V4.1 Flash $0.67, GLM 5.3 $5.72. Measured on real coding-agent traffic through the Standard Compute gateway (Sep 13 to Oct 12, 2026), priced with each request's own token counts and cache reads.

What is the best model for Kilo Code on a $19, $49 or $99 monthly budget?

For a month of regular use (1,478 requests, the median in our data), the highest-rated model that fits under $19: Claude Haiku 5.5 (about $17.73); under $49: Claude Haiku 5.5 (about $17.73); under $99: Claude Sonnet 5.5 (about $77.42). Heavier months cost proportionally more.

Every model on this page runs on Standard Compute

Standard Compute

Current frontier models Claude Fable 5.1 · GPT-6 Astra