# OpenCode Zen vs Anthropic Direct - Pricing Comparison ## What You Need to Know If you're using Anthropic's Claude models with OpenCode, you have two options: **OpenCode Zen** (curated gateway) or **direct Anthropic API**. The good news: **they cost exactly the same** for Claude models. OpenCode Zen charges the same per-token rates as Anthropic's API, with no markup beyond credit card processing fees. ## TL;DR - Should I Use Zen? **Use OpenCode Zen if:** - You want tested/verified model configurations - You need access to multiple providers (Chinese models, Grok, etc.) - You prefer pay-as-you-go with no billing account setup - You want team management features (coming soon: free workspaces) **Use Anthropic Direct if:** - You only use Claude models - You already have Anthropic API credits - You prefer direct provider relationship - Your organization requires vendor contracts **Bottom line:** For Anthropic-only users, it's mostly a preference. Prices are identical. ## Claude Model Pricing (Identical on Both) ### Current Models (Per 1M Tokens) | Model | Input | Output | Cached Write | Cached Read | |-------|--------|--------|--------------|-------------| | **Claude Sonnet 4.5** (≤ 200K) | $3.00 | $15.00 | $3.75 | $0.30 | | **Claude Sonnet 4.5** (> 200K) | $6.00 | $22.50 | $7.50 | $0.60 | | **Claude Sonnet 4** (≤ 200K) | $3.00 | $15.00 | $3.75 | $0.30 | | **Claude Sonnet 4** (> 200K) | $6.00 | $22.50 | $7.50 | $0.60 | | **Claude Haiku 4.5** | $1.00 | $5.00 | $1.25 | $0.10 | | **Claude Haiku 3.5** | $0.80 | $4.00 | $1.00 | $0.08 | | **Claude Opus 4.1** | $15.00 | $75.00 | $18.75 | $1.50 | ### Legacy Models | Model | Input | Output | Cached Write | Cached Read | |-------|--------|--------|--------------|-------------| | **Claude Opus 4** | $15.00 | $75.00 | $18.75 | $1.50 | | **Claude Sonnet 3.7** | $3.00 | $15.00 | $3.75 | $0.30 | | **Claude Opus 3** | $15.00 | $75.00 | $18.75 | $1.50 | | **Claude Haiku 3** | $0.25 | $1.25 | $0.30 | $0.03 | **Source:** Verified identical on both [OpenCode Zen pricing](https://opencode.ai/docs/zen/#pricing) and [Anthropic API pricing](https://claude.com/pricing) as of November 2025. ## What Zen Adds Beyond Claude ### Free Models (Exclusive to Zen) - **Grok Code Fast 1** - Free (limited time, data may be used for training) - **Code Supernova** - Free (limited time, stealth model beta) ### Chinese AI Models | Model | Input | Output | Cached Write | Cached Read | |-------|--------|--------|--------------|-------------| | **GLM 4.6** | $0.60 | $2.20 | - | $0.10 | | **Kimi K2** | $0.60 | $2.50 | - | $0.36 | | **Qwen3 Coder 480B** | $0.45 | $1.50 | - | - | ### OpenAI Models (via Zen) | Model | Input | Output | Cached Read | |-------|--------|--------|-------------| | **GPT 5** | $1.25 | $10.00 | $0.125 | | **GPT 5 Codex** | $1.25 | $10.00 | $0.125 | ## Real-World Cost Examples ### Example 1: Daily Coding Session **Scenario:** 2 hours of coding with Sonnet 4.5, typical file context - Input: ~150K tokens (codebase context) - Output: ~20K tokens (code generation) - **Cost:** ~$0.75 per session **Monthly (20 workdays):** ~$15/month ### Example 2: Heavy Context Work **Scenario:** Large codebase refactoring (> 200K tokens context) - Input: 300K tokens (large context window) - Output: 50K tokens (extensive changes) - **Cost:** ~$2.55 per session **Why the jump?** Pricing doubles for prompts > 200K tokens on Sonnet 4.5. ### Example 3: Budget-Conscious Development **Scenario:** Use Haiku 4.5 for routine tasks, Sonnet for complex work - 80% tasks on Haiku: $0.15/session avg - 20% tasks on Sonnet: $0.75/session avg - **Blended cost:** ~$0.27/session **Monthly (20 workdays):** ~$5.40/month ### Example 4: Team Environment (5 developers) **Scenario:** Mixed usage across team - 100 sessions/month total (varied models) - Average session: $0.50 - **Total:** $50/month for entire team **With Zen Team Plan:** - Standard seats: $25/person/month ($125 total) + API usage ($50) = $175/month - Includes admin controls, SSO, billing management ## Hidden Costs to Consider ### Credit Card Processing Fees **OpenCode Zen:** Passes through credit card fees at cost (typically 2.9% + $0.30) - $100 in API usage = ~$2.90 in fees - Total: $102.90 **Anthropic Direct:** Included in pricing (no separate fee disclosure) ### Minimum Commitments **OpenCode Zen:** - Individual: No minimum - Team: 5 members minimum ($125/month for standard seats) **Anthropic Direct:** - Free tier available - Pay-as-you-go with no minimum ### Batch Processing Discount **Both services offer 50% off** for asynchronous batch processing - Useful for non-realtime workloads - Example: Sonnet 4.5 input drops from $3/MTok to $1.50/MTok ## Privacy and Data Retention ### OpenCode Zen - **Zero-retention policy** (except during free model betas) - **Exceptions:** - Grok Code Fast 1: Data may be used for training during free period - Code Supernova: Data may be used for training during free period - OpenAI APIs: 30-day retention per OpenAI policy - Anthropic APIs: 30-day retention per Anthropic policy ### Anthropic Direct - **30-day retention** for all API requests - **No training** on API data per Anthropic's policy - Data stored in US **Both are equivalent** for Claude models - same 30-day retention policy. ## Advanced Features Comparison ### Prompt Caching (Both Services) Save on repeated context by caching prompts: - **Write cost:** 1.25x input price (one-time) - **Read cost:** 10x cheaper than full input - **TTL:** 5 minutes (standard), extended caching available **Example savings:** - Full input: 100K tokens at $3/MTok = $0.30 - Cached read: 100K tokens at $0.30/MTok = $0.03 - **Savings:** 90% on repeated context ### Service Tiers (Anthropic Direct Only) - **Priority:** Guaranteed availability, faster response (contact sales) - **Standard:** Default tier (same as Zen) - **Batch:** 50% discount for async workloads (also on Zen) OpenCode Zen doesn't explicitly offer priority tiers yet. ### Extended Tools (Anthropic Direct) **Web Search:** - $10 per 1K searches (doesn't include token costs) - Give Claude access to current web info **Code Execution:** - 50 free hours/day per organization - $0.05/hour beyond that - Python sandbox for data analysis **Note:** These features are available through Anthropic Direct API, not yet integrated in OpenCode Zen. ## Team Features ### OpenCode Zen Workspaces (Beta - Free) - **Roles:** Admin vs Member permissions - **Model access controls:** Enable/disable models org-wide - **Spending limits:** Set monthly caps per member - **Bring your own key:** Use existing Anthropic/OpenAI keys - **Billing:** Centralized billing management **Pricing:** Free during beta (pricing TBA) ### Anthropic Enterprise - SSO and domain capture - SCIM provisioning - Audit logs - Compliance API - Custom data retention **Pricing:** Contact sales **Comparison:** Anthropic Enterprise is more mature for large orgs with compliance needs. Zen workspaces are simpler and currently free. ## When to Switch ### From Zen to Anthropic Direct **Switch if:** - You only use Claude models exclusively - You need enterprise compliance features (SOC2, HIPAA, etc.) - You prefer vendor relationship with Anthropic directly - Your org already has Anthropic contracts ### From Anthropic Direct to Zen **Switch if:** - You want to experiment with other models (Chinese AI, GPT, Grok) - You prefer unified billing across multiple providers - You like tested/verified model configurations - You want simpler team management (once workspaces mature) ## Practical Recommendations ### For Individual Developers **If primarily using Claude:** Either works - pick based on preference - Zen: Easier experimentation with other models - Direct: Simpler if Claude-only **If exploring multiple models:** Use Zen - Access to Qwen, Kimi, GLM, Grok in one place - Free tier models available ### For Small Teams (2-10 devs) **Use Zen if:** - You want centralized billing - Team uses mixed models - You need basic access controls **Use Anthropic Direct if:** - Claude-only workflow - No need for team management features - Existing Anthropic relationship ### For Enterprises (10+ devs) **Use Anthropic Direct if:** - Compliance requirements (SOC2, HIPAA, etc.) - Need enterprise SLAs - Require audit logs and SCIM **Consider Zen if:** - Multi-provider strategy (Claude + others) - Want to test different model/provider combinations - OpenCode is primary coding interface ## Cost Optimization Tips ### 1. Use Prompt Caching Aggressively ```bash # Example: Large codebase as cached context # First call: 200K tokens write = $0.75 # Next 20 calls (within 5min): 200K tokens read = $0.06 each # Savings: $0.69 per call × 20 = $13.80 saved ``` ### 2. Right-Size Your Model | Task Type | Recommended Model | Cost/Session | |-----------|-------------------|--------------| | Code review | Haiku 4.5 | $0.10 | | Bug fixing | Haiku 4.5 or Sonnet 4.5 | $0.10-$0.50 | | Architecture design | Sonnet 4.5 | $0.50-$1.00 | | Complex refactoring | Sonnet 4.5 or Opus 4.1 | $1.00-$3.00 | ### 3. Use Batch Processing for Non-Urgent Work ```bash # Regular: 1M input tokens on Sonnet 4.5 = $3.00 # Batch: Same request = $1.50 (50% off) ``` **Good for batch:** - Test generation - Documentation writing - Code analysis - Non-interactive refactoring ### 4. Monitor Context Window Threshold **Sonnet 4.5 pricing doubles at 200K tokens:** - Keep context under 200K when possible - Use selective file inclusion (`@file.ts` strategically) - Consider splitting large refactors into smaller sessions ### 5. Try Free Models for Exploration On OpenCode Zen: - Use **Grok Code Fast 1** (free) for experimentation - Use **Code Supernova** (free) for testing workflows - Switch to paid models for production work ## Billing and Payment ### OpenCode Zen - **Model:** Prepaid credits - **Top-up:** Add credits as needed - **No subscription** for individual use - **Team plans:** $25-$150/person/month (minimum 5 members) ### Anthropic Direct - **Model:** Pay-as-you-go (billed monthly) - **Free tier:** Available with rate limits - **No commitment** required - **Enterprise:** Custom contracts ## Final Recommendation ### For You (Mainly Using Anthropic) Since you're primarily using Anthropic models, **either option works perfectly well** - the pricing is identical. **Go with OpenCode Zen if:** - You like the convenience of one login for everything OpenCode-related - You want to try other models (Qwen, Kimi, free Grok) without extra setup - You appreciate that Zen models are pre-tested/configured by the OpenCode team **Stick with Anthropic Direct if:** - You prefer managing API keys directly with the provider - You want a direct relationship with Anthropic (for support, enterprise features, etc.) - You're already comfortable with your current Anthropic setup **Cost-wise, there's no difference.** The choice is purely about convenience and feature access. ### Quick Setup Comparison **OpenCode Zen:** ```bash opencode auth login # Select: opencode # Visit: https://opencode.ai/auth # Add billing → Copy API key → Paste ``` **Anthropic Direct:** ```bash opencode auth login # Select: anthropic # Visit: https://console.anthropic.com # Create API key → Paste ``` Both take about 2 minutes. ## Related Resources - [[OpenCode - Managing and Searching Conversations]] - Session management tips - [OpenCode Zen Documentation](https://opencode.ai/docs/zen/) - [Anthropic API Pricing](https://claude.com/pricing) - [OpenCode Providers Guide](https://opencode.ai/docs/providers/) ## Updates and Changes **Last verified:** November 8, 2025 **Watch for:** - OpenCode Zen workspace pricing (currently free beta) - New models added to Zen catalog - Anthropic API price changes - Free model program deadlines (Grok, Supernova) Check official pricing pages for latest rates: - Zen: https://opencode.ai/docs/zen/#pricing - Anthropic: https://claude.com/pricing