TheRouter.aiTheRouter.ai
Back to Models

Claude Opus 4.7

anthropicanthropic/claude-opus-4.7

Claude Opus 4.7 is Anthropic's most capable model — 13% coding lift over Opus 4.6, tripled image resolution (2576px / 3.75 MP), new xhigh effort level, and task budgets for autonomous agent loops. Best for coding, agents, enterprise workflows, cybersecurity, and financial analysis.

Anthropic announcement ↗Anthropic model docs ↗Adaptive thinking & effort ↗

Claude Opus 4.7 is Anthropic's most capable generally available model, released April 16, 2026. Anthropic positions it as a coding- and agent-focused upgrade over Opus 4.6: 87.6% on SWE-bench Verified (up from 80.8%), 64.3% on SWE-bench Pro (up from 53.4%), and roughly 14% fewer steps on long-horizon agent workflows.

From a TheRouter operator's perspective, three things actually change. Vision input now accepts images up to 2,576 px on the long edge (about 3.75 MP), more than 3x prior Claude resolution. A new xhigh reasoning effort sits between high and max, intended as the Claude Code default for complex coding work. And the tokenizer was updated — Anthropic warns the same prompt may cost 1.00x to 1.35x more tokens than on Opus 4.6, so measure spend on real traffic before locking budgets.

Best for
  • • Long-running coding agents (multi-file refactors, repo-wide bug fixes) where Opus 4.6 stalled or hit tool-error loops
  • • Code review workloads — partner CodeRabbit reports >10% recall improvement on hard-to-detect bugs without losing precision
  • • Cybersecurity and document analysis pipelines that need high-resolution vision (2,576 px long edge)
  • • Enterprise workflows that benefit from multi-agent coordination — Opus 4.7 can orchestrate parallel sub-tasks instead of strict sequential execution
Reach for something else if
  • • Latency-sensitive chat — at xhigh and max effort the model thinks deeply; for sub-second responses route to Claude Haiku 4.5 or Claude Sonnet 4.6
  • • Cost-bounded high-volume workloads — Opus 4.7 is $6 in / $30 out per 1M tokens on TheRouter; for cheap reasoning-good completions route to DeepSeek V3.2 or Gemini 2.5 Flash
  • • Manual extended thinking with budget_tokens — no longer supported on Opus 4.7; you must use adaptive thinking via the effort parameter
Context Length
1M
Max Output
128K
Input Priceper 1M tokens
$6.00/ 1M tokens
Output Priceper 1M tokens
$30.00/ 1M tokens

Modalities

textimagepdf→text

Pricing Breakdown

TypeRate
Input$6.00 / 1M tokens
Output$30.00 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_ptoolstool_choiceresponse_formatreasoningstop

Specifications

Release date2026-04-16www.anthropic.com ↗verified
Anthropic model idclaude-opus-4-7docs.anthropic.com ↗verified
Training cutoffNot publicly disclosedunknown
Reasoning effort levelslow, medium, high (API default), xhigh (Claude Code default), maxplatform.claude.com ↗verified
Max vision input resolution2,576 px on long edge (~3.75 megapixels)www.anthropic.com ↗verified
Tokenizer change vs Opus 4.6Updated tokenizer; same prompt may use 1.00x–1.35x more tokenswww.anthropic.com ↗verified
DeploymentAnthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, GitHub Copilotaws.amazon.com ↗verified
LicenseAnthropic Usage Policy (proprietary, API-only)verified

Benchmarks

Full benchmarks page →
BenchmarkDistributionScoreSource
SWE-bench Verified
Up from 80.8% on Opus 4.6. Includes memorization-screen adjustments.
87.6%%www.anthropic.com ↗
SWE-bench Pro
Up from 53.4% on Opus 4.6; GPT-5.4 scores 57.7%, Gemini 3.1 Pro 54.2%.
64.3%%www.anthropic.com ↗
Terminal-Bench 2.0
Up from 65.4% on Opus 4.6.
69.4%%www.anthropic.com ↗
GPQA Diamond
Frontier models have effectively saturated this benchmark — GPT-5.4 Pro 94.4%, Gemini 3.1 Pro 94.3%.
94.2%%www.anthropic.com ↗
MCP-Atlas
Multi-tool MCP coordination benchmark.
77.3%%www.anthropic.com ↗

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL
curl https://api.therouter.ai/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "anthropic/claude-opus-4.7",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

API guide

Full API reference →

Chat completion

Standard chat call. TheRouter exposes Anthropic models behind the OpenAI-compatible /v1/chat/completions surface.

cURL
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-opus-4.7",
    "messages": [{"role": "user", "content": "Refactor this Python function for clarity."}]
  }'

More from anthropic

anthropic/claude-opus-4.6Previous Opus generation — cheaper, still strong, and the apples-to-apples baseline for every 4.7 benchmark.anthropic/claude-sonnet-4.6Mid-tier Anthropic model — better speed/cost balance for high-volume coding and chat.anthropic/claude-haiku-4.5Anthropic's fast, low-cost model — route here for latency-sensitive flows where Opus is overkill.anthropic/claude-opus-4.5Two generations back; useful for regression A/Bs and reproducibility of older traces.

Similar models

Cross-provider sibling models
openai/gpt-5.4
OpenAI's competitive flagship; Anthropic reports Opus 4.7 leads on SWE-bench Pro and Finance Agent.
openai/gpt-5-pro
OpenAI's deep-reasoning tier — compare for math-heavy and proof-style workloads.
google/gemini-2.5-pro
Google's flagship; per Anthropic's release notes, Gemini leads on BrowseComp and MMMLU multilingual Q&A.
deepseek/deepseek-v3.2
Open-weight alternative — DeepSeek V3.2 with sparse attention; much cheaper, comparable on many reasoning tasks.

News & changes

2026-04-16

Anthropic launches Claude Opus 4.7 with xhigh effort and 3.75 MP vision

TheRouter routed Opus 4.7 traffic from day one. The headline change for operators is the new xhigh reasoning level (default in Claude Code), tripled image resolution, and the updated tokenizer — same prompt may cost 1.00x to 1.35x more tokens than on 4.6, so re-baseline cost dashboards before locking quotas.

re-authored by TheRouteranthropic.com/news ↗
2026-04-16

Claude Opus 4.7 available in Amazon Bedrock at launch

AWS shipped Opus 4.7 on Bedrock the same day as the Anthropic API. For teams routed through TheRouter, this means a Bedrock-hosted fallback path is live without additional wiring — useful for VPC-bound workloads that can't egress to the Anthropic API directly.

re-authored by TheRouteraws.amazon.com/blogs/aws ↗
2026-04-16

Anthropic deprecates manual budget_tokens — adaptive thinking is now mandatory on Opus 4.7

If your existing Opus 4.6 client passes `thinking: {type: 'enabled', budget_tokens: N}`, that path is gone on 4.7. Migrate to `reasoning.effort` with xhigh or max. TheRouter forwards the new field unchanged; older request shapes will be rejected upstream.

re-authored by TheRouterplatform.claude.com ↗

Frequently asked

More FAQ →
How much better is Opus 4.7 at coding than Opus 4.6?

On Anthropic's own numbers: SWE-bench Verified 87.6% vs 80.8% (+6.8 points), SWE-bench Pro 64.3% vs 53.4% (+10.9 points), Terminal-Bench 2.0 69.4% vs 65.4%. Partner Rakuten reports 3x more production tasks resolved on their internal SWE-Bench; CodeRabbit reports +10% recall on hard-to-detect bugs.

www.anthropic.com ↗
Do I need to change my API code to use Opus 4.7?

Two breaking changes. (1) Manual extended thinking via `thinking: {type:'enabled', budget_tokens:N}` is removed — switch to `reasoning.effort` (low/medium/high/xhigh/max). (2) The tokenizer changed; the same prompt may consume 1.00x to 1.35x more tokens. Re-baseline cost and quota dashboards before flipping production traffic.

platform.claude.com ↗
When should I use xhigh vs max effort?

Anthropic's guidance: xhigh for advanced coding and complex agent loops (default in Claude Code), high for most intelligence-sensitive workloads, max only when evals show measurable headroom past xhigh. max often overthinks — most 'Opus 4.7 feels slow' reports trace back to running max by reflex.

platform.claude.com ↗
What image resolutions does Opus 4.7 accept?

Up to 2,576 px on the long edge, about 3.75 megapixels — more than 3x prior Claude models. This is a model-level change applied automatically; no API flag required. XBOW reports a 98.5% visual-acuity benchmark (vs 54.5% on Opus 4.6).

www.anthropic.com ↗
Where can I run Opus 4.7?

Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, GitHub Copilot (Pro+, Business, Enterprise) — and through TheRouter on the OpenAI-compatible /v1/chat/completions surface as model id 'anthropic/claude-opus-4.7'.

aws.amazon.com ↗
Fact ledger — every claim on this page traces here
sourceURLretrieved
Release datewww.anthropic.com ↗2026-05-22verified
Anthropic model iddocs.anthropic.com ↗2026-05-22verified
Training cutoff——unknown
Reasoning effort levelsplatform.claude.com ↗2026-05-22verified
Max vision input resolutionwww.anthropic.com ↗2026-05-22verified
Tokenizer change vs Opus 4.6www.anthropic.com ↗2026-05-22verified
Deploymentaws.amazon.com ↗2026-05-22verified
License——verified
SWE-bench Verifiedwww.anthropic.com ↗2026-05-22verified
SWE-bench Prowww.anthropic.com ↗2026-05-22verified
Terminal-Bench 2.0www.anthropic.com ↗2026-05-22verified
GPQA Diamondwww.anthropic.com ↗2026-05-22verified
MCP-Atlaswww.anthropic.com ↗2026-05-22verified
Finance Agentwww.anthropic.com ↗2026-05-22verified
Humanity's Last Exam (with tools)www.anthropic.com ↗2026-05-22verified
XBOW visual acuitywww.anthropic.com ↗2026-05-22verified
Anthropic launches Claude Opus 4.7 with xhigh effort and 3.75 MP visionanthropic.com/news ↗2026-05-22verified
Claude Opus 4.7 available in Amazon Bedrock at launchaws.amazon.com/blogs/aws ↗2026-05-22verified
Anthropic deprecates manual budget_tokens — adaptive thinking is now mandatory on Opus 4.7platform.claude.com ↗2026-05-22verified
How much better is Opus 4.7 at coding than Opus 4.6?www.anthropic.com ↗2026-05-22to verify
Do I need to change my API code to use Opus 4.7?platform.claude.com ↗2026-05-22to verify
When should I use xhigh vs max effort?platform.claude.com ↗2026-05-22to verify
What image resolutions does Opus 4.7 accept?www.anthropic.com ↗2026-05-22to verify
Where can I run Opus 4.7?aws.amazon.com ↗2026-05-22to verify
Customer Support