Anthropic's Claude Opus 4.8 Hits GitHub Copilot at a Steep 15x Cost

Claude Opus 4.8 is now generally available inside GitHub Copilot. Anthropic's latest flagship model brings meaningful gains in agentic coding, large-codebase navigation, and self-reported code reliability, and it lands at the same API price as Opus 4.7. The catch: it carries a 15x premium request multiplier in Copilot, making it the most expensive model in the picker by a wide margin.

What actually changed under the hood

Opus 4.8 builds on Opus 4.7 with improvements across benchmarks and is described by Anthropic as a more effective collaborator. The headline numbers back that up. On SWE-bench Verified, the model scores 88.6%, up from 87.6% on Opus 4.7 and 80.8% on Opus 4.6. On the harder SWE-bench Pro benchmark, it hits 69.2%, up from 64.3%. SWE-bench Pro is a benchmark that tests whether a model can autonomously resolve real GitHub issues, making it a strong proxy for practical coding ability.

Opus 4.8 is also the strongest computer-use and browser-agent model Anthropic has tested, scoring 84% on Online-Mind2Web, a meaningful jump over both Opus 4.7 and GPT-5.5. That said, Terminal-Bench 2.1 for agentic terminal coding still belongs to GPT-5.5 at 78.2%, with Opus 4.8 coming in at 74.6%.

The reliability story is the real headline

Raw benchmark scores are only part of the picture. The more interesting shift is in what Anthropic calls "honesty" -- the model's tendency to flag its own mistakes rather than silently ship broken code into your pipeline.

• Anthropic's evaluations found the model to be around four times less likely than its predecessor to leave flaws in its own code unremarked.
• Early testers found Opus 4.8 sharper in judgment when performing agentic tasks, more likely to flag uncertainty, and less likely to make unsupported claims.
• Opus 4.8 also demonstrated substantially lower rates of misaligned behavior compared to Opus 4.7.
• It also fixes the comment-verbosity and tool-calling issues that were observed with Opus 4.7.

This matters most in long-running agentic sessions, where a model that silently passes a flawed function can cascade errors across dozens of downstream steps before anyone notices. This is the kind of thing that doesn't show up in a single demo but compounds across long-running agentic sessions.

New features shipping alongside the model

Anthropic also announced a major new research preview feature called dynamic workflows in Claude Code. This feature allows Claude to plan large tasks and run hundreds of parallel subagents in a single session, targeting large-scale engineering work like codebase migrations.

Two other features are worth knowing about if you build agentic pipelines:

• Effort control: Users on claude.ai now have control over the amount of effort Claude puts into a task. Configurable levels range from low to max, letting you trade response depth for speed and cost.
• Mid-conversation system messages: You can now send a system message immediately after a user turn in the messages array, appending updated instructions later in a long-running conversation without restarting the full system prompt. This preserves prompt cache hits on earlier turns and reduces input cost on agentic loops.

Fast mode changes the cost math

The most underrated part of this release is what happened to fast mode pricing. Fast mode for Opus 4.8 can run at up to 2.5 times the speed and is now priced at $10 per million input tokens and $50 per million output tokens, making it three times cheaper than previous fast mode models. Hanlin Tang, CTO at Neural Networks, stated that Opus 4.8 unlocks a step change in agentic reasoning for Databricks' Genie, handling complex, multi-step questions faster and at a 61% lower token cost for PDF and diagram reasoning.

Who can use it, and what it costs in Copilot

Claude Opus 4.8 is available to Copilot Pro+, Business, and Enterprise users. You can select the model in the model picker in Visual Studio Code in all modes (chat, ask, edit, and agent), with rollout being gradual. It's also available in Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, GitHub Mobile, and github.com.

The pricing situation in Copilot deserves a close read. The model launched with a 15x premium request multiplier until usage-based billing launched on June 1, 2026. Base plan prices are unchanged ($10 Pro, $39 Pro+, $19 Business, $39 Enterprise per month), with premium request units being replaced by GitHub AI Credits tied to token consumption.

For Enterprise and Business admins: administrators have to enable the Claude Opus 4.8 policy in Copilot settings before developers can pick the model. As of late May, Enterprise owners can scope model availability per organization rather than enterprise-wide, meaning a security review team can get Opus 4.8 while a frontend team stays on Auto.

When to actually reach for it

At 15x the cost of a standard request, Opus 4.8 in Copilot is not a model you use for every autocomplete. The right mental model is to treat it like a senior engineer you call in for specific problems.

• Large-codebase refactors and migrations: Early testing shows it handles complex problem-solving and large-codebase navigation with notable improvement over previous versions.
• Security-sensitive code review: Opus 4.8 is around four times less likely than Opus 4.7 to let code flaws pass unremarked. For a security team running review passes through the cloud agent, that delta is the entire point of paying 15x.
• Long-horizon agentic tasks: Anthropic describes it as its most capable model for complex reasoning, long-horizon agentic coding, and high-autonomy work.
• Skip it for: Inline completions, chat questions about syntax, and short cloud agent tasks that finish in three or four steps. Auto's router will pick a cheaper model for these by default and apply a 10% discount.

The bigger picture

Claude Opus 4.8 shows how frontier AI competition is moving beyond benchmark performance towards reliability in professional workflows. Features such as effort control, dynamic workflows, cheaper fast mode, and stronger agentic task performance point to a market shift in which AI systems are expected to manage longer, more complex work while giving users greater control over cost, speed, and reasoning depth.

GitHub's Copilot is itself evolving in parallel. GitHub Copilot now supports multiple AI models, each optimized for different workflows, from quick edits to deep debugging to multi-step agentic tasks that generate or modify code across an entire repository. Opus 4.8 sits at the top of that stack, and if you're running the kind of work that justifies the cost, regular API pricing remains $5 per million input tokens and $25 per million output tokens -- unchanged from Opus 4.7. The upgrade is free in that sense. The question is whether your tasks are complex enough to earn it.