Coding agent cost per task: Artificial Analysis Coding Agent Index 2.0
What it costs, in API tokens, for a coding agent to finish one benchmark task, from $0.16 to $13.30 depending on the agent and model.
Key takeaways
- Top score: Claude Code with Opus 5.5 (max) scores 61.7 at $11.42 per task.
- Best value near the top: Codex with GPT-6.1 Sol (max) scores 54.1 at $1.50 per task, about 88% of the top score for about 13% of the cost.
- Cheapest strong option: DeepSeek Harness with DeepSeek V4.1 Flash 1017 scores 46.3 at $0.30, and Codex with DeepSeek V4.1 Flash scores 47.3 at $0.39.
- The spread is huge: the most expensive configuration (Claude Code with Sonnet 5.5 max, $13.30) costs about 83 times the cheapest (Codex with GPT-6 Luna max, $0.16).
- The harness matters, not just the model: the same model can cost very different amounts in different agents because of how many steps they take and how well they use prompt caching.
Score and cost per task by agent and model
Sorted by index score. "Points per $" is index score divided by cost per task (our calculation).
| Agent | Model (effort) | Index score | Cost per task | Points per $ |
|---|---|---|---|---|
| Claude Code | Opus 5.5 (max) | 61.7 | $11.42 | 5.4 |
| Devin Fusion CLI | Claude Opus 5.5 Max + GPT-6 Luna High | 61.5 | $8.00 | 7.7 |
| Claude Code | Sonnet 5.5 (max) | 59.3 | $13.30 | 4.5 |
| Devin Fusion CLI | Claude Fable 5.1 XHigh + SWE-2 Medium | 58.7 | $8.20 | 7.2 |
| Claude Code | Fable 5.1 (max, with fallback) | 57.6 | $11.53 | 5.0 |
| Codex | GPT-6 Astra (max) | 54.7 | $7.04 | 7.8 |
| Codex | GPT-6.1 Sol (max) | 54.1 | $1.50 | 36.1 |
| Antigravity CLI | Gemini 4 Argon (high) | 53.2 | $5.86 | 9.1 |
| Devin Fusion CLI | GPT-6 Astra XHigh + SWE-2 Medium | 53.1 | $4.34 | 12.2 |
| Muse Code | Muse Wonder (max) | 52.0 | $8.05 | 6.5 |
| Opencode | GLM-5.3 (max) | 48.9 | $5.23 | 9.3 |
| Codex | DeepSeek V4.1 Flash (max) | 47.3 | $0.39 | 121.3 |
| Grok Build | Grok 4.7 (xhigh) | 46.6 | $9.48 | 4.9 |
| DeepSeek Harness | DeepSeek V4.1 Flash 1017 (max) | 46.3 | $0.30 | 154.3 |
| Muse Code | Muse Spark 1.3 (max) | 45.7 | $3.91 | 11.7 |
| Claude Code | Step 5 Preview | 43.1 | $1.84 | 23.4 |
| Kimi Code CLI | Kimi K3 | 41.4 | $5.17 | 8.0 |
| Claude Code | Qwen3.8 Max | 39.2 | $3.80 | 10.3 |
| Claude Code | Haiku 5.5 (max) | 37.3 | $2.67 | 14.0 |
| Antigravity SDK | Gemini 3.8 Flash (high) | 33.3 | $2.51 | 13.3 |
| Codex | GPT-6 Luna (max) | 31.1 | $0.16 | 194.4 |
| Muse Code | Muse Spark Open (xhigh) | 26.3 | $3.54 | 7.4 |
What the numbers mean
Coding Agent Index v2.0 is the equal-weight average of three benchmarks: FrontierCode v1.1, Terminal-Bench 4.0 and SWE-Atlas-QnA-AA. Cost per task is the average pay-per-token API cost per task, including cached-input discounts and cache-write charges where providers bill them. It is not a subscription price: many people run these agents on $20/month plans instead. Infrastructure and engineering time are not included.
Source: Artificial Analysis, AI Coding Agent Benchmarks & Leaderboard (data as of Oct 22, 2026) and Coding Agent Index v2.0 methodology. Scores shown on a 0-100 scale, costs rounded to the cent. TokenTab is not affiliated with Artificial Analysis.
Estimate your own agent's cost
Benchmarks are a guide; your prompts, tools and task mix decide your bill. Plug your own token counts and tool calls into the calculator to see the cost per task, per day and per month on every major model.