Anthropic released Claude Opus 4.7 on April 16, 2026, raising its SWE-bench Verified coding score to 87.6% from 80.8% for Opus 4.6 while holding pricing flat at $5 per million input tokens and $25 per million output tokens.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source 3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source The model shipped the same day into GitHub Copilot for paid tiers and added self-verification, a new xhigh effort level, and roughly 3.3 times higher vision resolution.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source 2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source The stake is whether coding leadership at flat pricing is a defensible moat in the enterprise developer market. We assess with moderate confidence that Opus 4.7 strengthens Anthropic's position among enterprise developers, but that a tokenizer change and rising Copilot request multipliers mean the flat headline price is not the flat effective price.

What the release is engineered to do

The gains are concentrated in software engineering, the segment Anthropic has made its priority. SWE-bench Verified, a curated set of 500 human-validated GitHub issues, rose 6.8 percentage points to 87.6%, and a vision benchmark climbed from 77.4% to 91.0%, while the context window stayed at 1 million input tokens.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source The feature set reads as reliability engineering for agents: self-verification lets the model check its own outputs before reporting, an xhigh effort level sits between high and max for a better reasoning-to-latency tradeoff, and file-system memory improves persistence across multi-session work.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source Anthropic framed the target as handing off the hardest coding work that previously needed close supervision.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source Holding price while improving the benchmark is the core commercial signal: the pitch is more capability per dollar, aimed at buyers already standardized on Opus.

The distribution matters as much as the model. Opus 4.7 became generally available in GitHub Copilot on launch day for Pro+, Business, and Enterprise users, replacing earlier Opus versions in the model picker across editors from VS Code to JetBrains and Xcode.2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source That places the coding gains directly in the tool where enterprise developers already work, which is where a coding moat would actually be defended.

Second order effects and the ledger

The first order fact is a higher coding benchmark at the same list price. The second order effect is pressure on rivals to match capability without raising price, and pressure on Anthropic to prove that a benchmark lead converts into switching costs rather than a number that resets at the next release.

Who gains: Anthropic, which reinforces its identity as the enterprise coding model and uses Copilot distribution to reach developers at the point of work.2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source Enterprise developer teams gain measurably better performance on hard, long-running coding tasks at an unchanged sticker price.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source Who loses: competing frontier coding models, which now must beat 87.6% on SWE-bench Verified or undercut on price to stay in enterprise consideration.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source The caveat cuts against the clean flat-price story, and honesty requires naming it: the new tokenizer means the same text can consume 1.0 to 1.35x more tokens than on Opus 4.6, so the effective cost per task can rise even though the per-token price did not move.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source In GitHub Copilot, the model launched with a 7.5x premium request multiplier during promotional pricing that rose to 15x when the promotion ended on May 1.2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source We assess with moderate confidence that the real-world cost to run Opus 4.7 rose for many workloads despite the unchanged headline rate, which weakens the flat-price framing.

The counter-case

The strongest argument against a durable moat is that coding benchmarks are the most contested and fastest-resetting scores in the field. A 6.8-point SWE-bench gain is real, but rivals iterate on the same benchmark, and a lead measured in single-digit percentage points tends to evaporate within a release cycle or two.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source If the moat rests on the number rather than on switching costs, it is shallow. The flat-pricing claim, the second pillar of the moat argument, is itself qualified by the tokenizer change and the Copilot multiplier increases, which mean the effective economics moved even as the list price did not.2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source 3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source The thesis that coding is a durable Anthropic moat fails if the next competing release matches 87.6% at lower effective cost and enterprises, facing rising per-task token consumption, treat models as interchangeable and route by price. What would have to be true for the moat to hold is that agentic reliability features such as self-verification and file-system memory create real switching friction, not just a higher benchmark.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source

What to watch

  • The SWE-bench lead survives the next rival release. If a competing model matches or beats 87.6% within two quarters, the coding lead is a cadence advantage, not a moat; if Anthropic holds it through the next cycle, the moat argument strengthens.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source
  • Effective cost per task, not list price. Watch whether enterprise buyers report rising per-task token spend from the new tokenizer over the next two quarters, which would undercut the flat-price story regardless of the unchanged headline rate.3 LLM Stats 2026-04-17 Opus 4.7 raised SWE-bench Verified from 80.8% to 87.6% and CharXiv-R vision from 77.4% to 91.0% at the same $5/$25 pricing, but a new tokenizer can consume 1.0 to 1.35x more tokens than Opus 4.6 for the same text. Open source
  • Copilot multipliers settle. The Copilot premium request multiplier moving from 7.5x to 15x after May 1 is the near-term cost signal; whether it falls back or holds will show the true cost of running Opus 4.7 in the tool where it is distributed.2 GitHub (github.blog) 2026-04-16 Opus 4.7 became generally available in GitHub Copilot on April 16, 2026 for Pro+, Business, and Enterprise, replacing earlier Opus versions, launching at a 7.5x premium request multiplier during promotional pricing that rose to 15x on May 1. Open source
  • Self-verification shows up in unsupervised handoffs. Named cases of teams handing hard coding work to Opus 4.7 without close supervision within two quarters would validate the reliability moat; their absence would suggest the gain stayed a benchmark, not a workflow change.1 Anthropic 2026-04-16 Claude Opus 4.7 released April 16, 2026 at unchanged pricing of $5 and $25 per million tokens, adding an xhigh effort level, over 3x higher vision resolution, self-verification of outputs, and improved file-system memory for multi-session agentic work. Open source The signal to track is not the launch score; it is whether the effective cost and reliability, taken together, actually lock enterprises in rather than merely topping a leaderboard for one cycle.