Claude Opus 5 API
Choose Claude Opus 5
Anthropic’s newest Opus-tier flagship for the hardest coding, long-running agents, and judgment-heavy review. Five billing dimensions — input, cache write, cache read, output, and the web search tool — are metered separately, with token pricing 5% below official rates.
Claude Opus 5
Anthropic Opus-tier flagship model
claude-opus-5Repository-scale software engineering, long-horizon agent runs with heavy tool use, high-stakes code review, and research or analysis work where a wrong answer costs more than a premium token rate.
Claude Opus 5 pricing
Estimate a request with the interactive pricing calculator. EvoLink token pricing is 5% below Anthropic’s official rates.
Token calculator
Enter the token mix and web searches for one request.Estimated request cost
Claude Opus 5Minimum charge: 0.01 credits per request.
EvoLink vs Anthropic direct
Same token mix, using the displayed EvoLink rates.Budget guide
Approximate requests using the current token mix.For quick testing
For regular development
For production evaluation
Model pricing
| Model | Context | Input tokens | Cache write tokens | Cache read tokens | Output tokens | Web searches |
|---|---|---|---|---|---|---|
Claude Opus 5claude-opus-5 | All context sizes | $4.750 / 1M-5% 323 cr / 1M$5.000official price | $5.939 / 1M-5% 403.8 cr / 1M$6.250official price | $0.475 / 1M-5% 32.3 cr / 1M$0.500official price | $23.750 / 1M-5% 1615 cr / 1M$25.000official price | $0.010 / search0.68 cr / search |
Claude Opus 5
All context sizesUSD and credits are shown per 1M tokens. Check the current rates in the pricing table before production use.
The web search tool is billed per search at the official rate, on top of the tokens the search results add to the request.
Claude Opus 5 API for the hardest coding and agent work
Call Anthropic’s newest Opus-tier flagship through EvoLink’s unified API at $4.75 / $23.75 per million input / output tokens — 5% below official pricing. Includes a 1,000,000-token context window, 128K max output, prompt caching with separate write and read rates, and the per-search web search tool.
Prompt caching billed as two separate dimensions
Claude prompt caching charges a write price when a prefix is stored and a much lower read price when it is reused. Stable system prompts, repository instructions, and tool schemas pay the write once and then read cheaply across a long agent run.
1M-token context with a 128K max output
Keep connected code, specifications, documents, and agent state in one working context. Treat the limit as capacity rather than a target: retrieve relevant evidence, cache the stable prefix, and set task-appropriate output budgets.
Thinking and effort change latency and token use
Anthropic enables adaptive thinking by default on Opus 5, and the Claude API defaults effort to high. Before production, verify parameter support on the EvoLink endpoint you use and measure the actual response, output tokens, and latency. Do not treat max_tokens as a thinking budget.
Replay real workloads before shifting production traffic
The same per-token rate does not guarantee the same completed-task cost. Model behavior and tokenizer changes can alter billed tokens, retries, and tool paths, so compare success rate, total cost, and human correction on production samples before shifting traffic.
Where Claude Opus 5 earns its premium rate
Opus-tier pricing only pays off when a task is hard enough that a better answer removes work downstream. Route to Opus 5 by the cost of being wrong, not by the cost per token.
Repository-scale software engineering
Inspect large codebases, trace dependencies across services, and plan multi-file changes while holding architectural constraints in one context. Measure accepted patches and review time rather than isolated snippet quality.
Long-horizon agents with heavy tool use
Multi-step orchestration, tool selection, and code execution over long runs, where a single wrong branch costs far more than the token difference. Replay complete assistant messages, tool-call IDs, arguments, and tool results across turns.
High-stakes review and analysis
Security review, migration planning, contract and research analysis, and anything where a confident wrong answer is expensive. The 1M-token window keeps the full evidence set in one place.
When another route is the better choice
Everyday coding, high-volume agents, classification, and latency-sensitive chat usually belong on Claude Sonnet 5. Claude Opus 4.8 carries a deeper discount on EvoLink and remains the cheaper Opus route when its answers are good enough.
Five billing dimensions, metered separately
Claude Opus 5 does not bill as a single blended rate. Understanding which dimension a workload actually consumes is what makes an Opus-tier budget predictable.
Input and output tokens
The two dimensions every request touches: $4.75 per 1M input tokens and $23.75 per 1M output tokens on EvoLink, 5% below Anthropic’s $5.00 and $25.00. Output dominates on generation-heavy work, so cap max_tokens by task complexity.
Cache write and cache read
Storing a prefix costs $5.939 per 1M tokens; reusing it costs $0.475 per 1M tokens. A stable system prompt, repository instructions, and tool schemas pay the write once and then read at roughly a tenth of the input rate for the rest of the run.
Web search tool
Billed per search at $0.01 — the official rate, with no discount applied. Search results also enter the request as ordinary input tokens, so a research-heavy turn pays both the per-search fee and the tokens it pulls in.
Adopt Claude Opus 5 without replacing your model stack
Keep one EvoLink client and API key, route the hardest workloads to Opus 5, and retain tested alternatives for simpler or latency-sensitive traffic.
Use the existing compatible client flow
Point the client at EvoLink, authenticate with the same API key, and select claude-opus-5. Both the OpenAI-compatible Chat Completions endpoint and the Anthropic Messages endpoint work.
Use the documented API model ID
The API route is claude-opus-5 on both protocols. The page URL and the model parameter are identical for this model.
Structure prompts for cache reuse
Put the stable system prompt, repository instructions, and tool schemas at the front of the request so the expensive prefix is written once and read cheaply for the rest of the session.
Route tasks by completed-task value
Compare success rate, retries, output tokens, cache hits, wall-clock time, and human correction against Claude Opus 4.8 and Sonnet 5 — not token price alone.
Copy the model ID and send your first request
This product page keeps only a verifiable minimal request. The full API guide covers thinking, effort, tool calls, failure handling, and production migration.
curl https://direct.evolink.ai/v1/messages \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-5",
"max_tokens": 8192,
"messages": [
{"role": "user", "content": "Review this pull request and list the blocking issues."}
]
}'Compare Claude routes after workload testing
EvoLinkFirst verify whether Opus 5 reduces retries and review effort on your own tasks. Then compare price, context, caching, and workload fit to choose the production route.
| Model | Claude Opus 5 | Claude Opus 4.8 | Claude Sonnet 5 |
|---|---|---|---|
| Input / output | $4.75 / $23.75 | $4.5 / $22.5 | $1.8 / $9 |
| Context | 1M | 1M | 1M |
| Caching | Read + write | Read + write | Read + write |
| Best for | Repository-scale software engineering, long-horizon agent runs with heavy tool use, high-stakes code review, and research or analysis work where a wrong answer costs more than a premium token rate. | The previous Opus flagship, still the cheaper Opus route on EvoLink for premium coding and agent work. | Opus-class coding and agent performance at a balanced Sonnet rate for everyday production traffic. |
Other Claude models on EvoLink
Claude Opus 4.8
The previous Opus flagship with a deeper EvoLink discount for workloads it can still complete reliably.
View model
Claude Sonnet 5
A more balanced route for everyday coding, high-volume agents, and latency-sensitive production traffic.
View model
Claude Fable 5
A higher-capability Claude route for the longest and most demanding agent workloads where quality outweighs cost.
View model
Claude Haiku 4.5
The fastest, lowest-cost Claude route for sub-agents, high-volume classification, and low-latency chat.
View modelOther text models
GPT-5.6
OpenAI’s tiered frontier family for routing across capability, latency, and cost.
View model
Grok 4.5
xAI’s reasoning and tool-use route with a 500K context window and server-side search tools.
View model
Kimi K3
A long-context reasoning model for repository-scale coding and multi-document work.
View model
Gemini 3.6 Flash
A fast, low-cost route for production traffic that does not need Opus-tier judgment.
View modelRelated reading
Claude Opus 6 Release Watch
Track the next Opus generation without treating its release date, API access, model ID, pricing, or specifications as confirmed.
Read guide
Claude Opus 5 vs Claude Opus 4.8
Which Opus route to run in production, and what has to improve for Opus 5 to be worth the smaller discount.
Read guide
Claude Opus 5 vs Claude Fable 5
When Opus 5 should be the premium default and when Fable 5 can justify twice the official token price.
Read guideClaude Opus 5 vs GPT-5.6
How the two frontier routes compare on pricing dimensions, context, and agent workloads.
Read guideGPT-6 vs Claude Opus 5 after Astra
Compare callable Opus 5 facts with unresolved Astra and GPT-6 claims without inventing specifications.
Read guide
Claude Opus 5 API Guide
Make your first request, configure thinking and effort, migrate safely, and use the production checklist.
Read guideClaude Opus 5 API FAQ
Is the Claude Opus 5 API available through EvoLink?
Yes. Claude Opus 5 is available on EvoLink as a production route under model ID "claude-opus-5".
What model ID should I use for the Claude Opus 5 API?
Send model "claude-opus-5". The same ID works for both the OpenAI-compatible Chat Completions endpoint and the Anthropic Messages endpoint.
How much does the Claude Opus 5 API cost on EvoLink?
EvoLink pricing is 5% below the official rate: $4.75 per 1M input tokens and $23.75 per 1M output tokens, against Anthropic’s $5.00 and $25.00.
How is Claude Opus 5 prompt caching billed?
Cache write and cache read are separate dimensions: $5.939 per 1M cache write tokens and $0.475 per 1M cache read tokens on EvoLink, versus $6.25 and $0.50 official.
How is the Claude Opus 5 web search tool billed?
Web search is charged per search at $0.01, the official rate with no discount applied. Tokens the search results add to the request are billed normally.
Which protocols and SDKs work with Claude Opus 5?
Use the OpenAI-compatible Chat Completions endpoint or the Anthropic Messages endpoint with the same EvoLink API key. Claude Code and other Anthropic-native clients work by pointing the base URL at EvoLink.
What context window and max output does Claude Opus 5 support?
The EvoLink route records a 1,000,000-token context window with up to 128K output tokens per response.
Is thinking enabled by default on Claude Opus 5?
Anthropic enables adaptive thinking by default on Opus 5, and the Claude API defaults effort to high. Confirm that the specific parameters you need pass through your selected EvoLink endpoint by checking the API docs and an actual response.
What workloads are a good fit for Claude Opus 5?
Claude Opus 5 fits especially demanding coding, long-running agents with heavy tool use, and critical reviews where errors create high downstream costs.
Can I use Claude Opus 5 alongside other model routes?
Yes. With the same EvoLink client and API key, route difficult tasks to Claude Opus 5 and keep tested alternatives for simpler or latency-sensitive requests.
What should a production Claude Opus 5 evaluation measure?
Track first-pass success, accepted deliverables, retries, output tokens, cache hit rate, valid tool calls, time to accepted result, human correction, and fallback rate — not the per-token rate alone.