Relace
Listed at https://relace.ai
Overall Rank #43 ⭐ Consider
⚠️ Overseas service with no mainland-China-optimized endpoint; China latency needs measurement. Code-related API defaults are not designed for mainland-China compliance scenarios. | 🌍 International
💰 Token Pricing
| Type | Price | Note |
|---|---|---|
| Input | 按每百万 token 计费（2026-11-02 OpenRouter 实时 Relace 端点）：glm-5.2 $0.0065、deepseek-v4.1-flash $0.035、deepseek-v4-flash $0.05、glm-5.3 $0.0703、glm-5.3-flash $0.0285、deepseek-v4-pro $0.90、kimi-k3 $0.1824、relace-apply-3 $0.85、relace-search $1.00 | per million tokens |
| Output | 按每百万 token 计费（2026-11-02 OpenRouter 实时 Relace 端点）：glm-5.3-flash $0.0923、deepseek-v4-pro $1.50、relace-apply-3 $1.25、relace-search $3.00、glm-5.3 $4.3843、deepseek-v4-flash $0.45、deepseek-v4.1-flash $0.50、glm-5.2 $7.9057、kimi-k3 $9.3388 | per million tokens |
💡 Free Credits: No traditional free token tier; free accounts are rate-limited to 15 requests per minute
🤖 Supported Models (12)
relace-search (own, BF16, 256K ctx, $1.00/$3.00, Fast Agentic Search)relace-apply-3 (own, FP8, 256K ctx, $0.85/$1.25, Instant Apply)deepseek-v4-pro (FP4, 1M ctx, $0.90/$1.50, cached $0.25)deepseek-v4-flash (FP4, 1M ctx, $0.05/$0.45, cached $0.05)deepseek-v4.1-flash (FP4, 1M ctx, $0.035/$0.50, cached $0.005)glm-5.3 (FP4, 1M ctx, $0.0703/$4.3843)glm-5.3-flash (FP4, 1M ctx, $0.0285/$0.0923)glm-5.2 (FP4, 1M ctx, $0.0065/$7.9057)kimi-k3 (FP4, 1M ctx, $0.1824/$9.3388, cached $0.18)
✨ Pros
- ✓Three agent-specific endpoints: Compact (trace compaction, >50k tok/s), Instant Apply (file merge, ~10k tok/s), Fast Agentic Search (agentic codebase search)
- ✓Homepage states 30 trillion tokens served in the last 30 days; top OpenRouter provider by daily tokens
- ✓Input-side floor pricing: Relace endpoints for deepseek-v4-flash, glm-5.3-flash and deepseek-v4-pro are the cheapest inputs on OpenRouter
- ✓GLM 5.2 at $0.0065/M input is the cheapest input tier on the platform
- ✓Zero Data Retention Policy for code generation models (effective 2026-09-21); KV cache never persists
- ✓Relace Repos offers fully managed version control for AI applications, with an MCP server for any coding agent
⚠️ Cons
- ×Publishes three mutually inconsistent price lists (marketing page 2026-10-29, hand-maintained docs table, live OpenRouter endpoints); same model ID differs by up to 4x
- ×Compact has no published standalone rate; cost accounting requires reading your own invoice
- ×No general-purpose reasoning model; the three products each solve one narrow problem
- ×Free tier is only 15 requests per minute, enough for evaluation
- ×GLM output rates are high (glm-5.2 output $7.9057/M is 5.6x Morph's), poor value for long-form generation
- ×Relace Repos binds source control and agent runtime to the same vendor
🎯 Best For
Teams that need long-running coding agents to be cost-feasible; source-code workloads that require zero data retention; batch pipelines that want the lowest input price for long context; and agent-infrastructure buyers willing to pay separately for agentic search, trace compaction and file merge