★ Reading this for free? Get 20 structured AI courses + per-chapter AI tutor — the first chapter of every course free, no card.Start free in 30 seconds

Anthropic · Closed-API Model · Current flagship

Claude Opus 4.8: Anthropic's Flagship, Tested

Claude Opus 4.8 (May 28, 2026) is Anthropic's current flagship and, as of June 2026, the #1 model on the Artificial Analysis Intelligence Index at 61.4 — narrowly ahead of GPT-5.5 (60.2) and its own predecessor Opus 4.7 (57.3). It scores 88.6% on SWE-Bench Verified and 69.2% on SWE-Bench Pro, keeps Opus 4.7's $5/$25 per-million-token pricing and 1M-token context, and adds an optional 2.5× Fast Mode plus an effort control. This review covers what changed from 4.7, how Fast Mode and effort work, the benchmark numbers, and which open-weight models come closest if you need to self-host.

📅 Published: June 19, 2026🔄 Last Updated: June 19, 2026✓ Manually Reviewed

Current flagship (June 2026): Opus 4.8 is Anthropic's top model. It replaced Claude Opus 4.7 (April 16, 2026), which is now the previous generation. Above the Opus tier, Anthropic's Mythos-class Fable 5 & Mythos 5 remains suspended worldwide by export-control order. Source: Anthropic's 4.8 announcement.

Note: Opus 4.8 is API-only — it cannot be downloaded or run locally. For frontier-class reasoning you can self-host, see DeepSeek V4-Pro, Kimi K2.6, and GLM-5.

Key takeaways

  • #1 Intelligence Index — 61.4 on Artificial Analysis, ahead of GPT-5.5 (60.2) and Opus 4.7 (57.3).
  • 88.6% SWE-Bench Verified · 69.2% SWE-Bench Pro — leads agentic coding (GPT-5.5 is 58.6% on Pro).
  • $5/$25 per Mtok — flat vs Opus 4.7; optional Fast Mode is $10/$50 for ~2.5× speed.
  • 1M context + effort control — effort defaults to high; extra/max for the hardest tasks.
  • API-only — for self-hosting, DeepSeek V4-Pro and Kimi K2.6 are the closest open-weight matches.

Quick verdict

Opus 4.8 is the model to default to if you want the single strongest closed model in June 2026 — it tops the Artificial Analysis Intelligence Index and leads agentic coding (69.2% SWE-Bench Pro). It's a drop-in upgrade over Opus 4.7 at the same $5/$25 price and 1M context, so there's little reason to stay on 4.7.

Where it loses: for most everyday coding, Claude Sonnet 4.6 ($3/$15) still wins on value — Opus 4.8 is worth its premium only on the hardest 5-10% of tasks. And it's API-only: for privacy-required or offline work, DeepSeek V4-Pro is the closest frontier open-weight you can self-host.

Opus 4.8 specs at a glance

VendorAnthropic
Release dateMay 28, 2026
StatusCurrent flagship
Model IDclaude-opus-4-8 · 1M variant claude-opus-4-8[1m]
Context window1,000,000 tokens (default on API, Bedrock, Vertex; 200K on Microsoft Foundry)
ModalitiesText · Code · Vision
Effort controlYes — defaults to high (extra/max available)
Fast ModeOptional — ~2.5× speed at $10/$50 per Mtok
LicenseProprietary
Local self-hostable?No

What changed from Opus 4.7?

Opus 4.8 landed 41 days after Opus 4.7 — a drop-in upgrade at the same list price and context. The gains show up across both real-world agentic coding and frontier reasoning, plus the new Fast Mode.

 Opus 4.8 (current)Opus 4.7 (previous)
ReleasedMay 28, 2026April 16, 2026
SWE-Bench Verified88.6%87.6%
SWE-Bench Pro69.2%64.3%
Artificial Analysis Index61.457.3
Context window1M tokens1M tokens
API pricing (in/out)$5 / $25 per Mtok$5 / $25 per Mtok
Fast Mode~2.5× at $10/$50— (pricier)

Beyond the scores, Anthropic highlights parallel-subagent dynamic workflows in Claude Code, mid-task system messages on the Messages API, and measurable honesty gains in its alignment assessment. The effort parameter and the rest of the platform features carry over from 4.7 unchanged. Sources: Anthropic news · Claude API docs.

Fast Mode & effort control

Opus 4.8 gives you two levers to trade speed, cost, and quality, and you can combine them per request.

Fast Mode — ~2.5× speed, $10/$50 per Mtok

An optional setting that runs Opus 4.8 at roughly 2.5× the standard output throughput for double the per-token price ($10/$50 vs $5/$25). Anthropic ships it about 3× cheaper than the equivalent fast mode on previous Claude models. Reach for it on latency-sensitive agents, interactive coding loops, and high-throughput pipelines where wall-clock time matters more than the lowest per-token cost.

Effort control — defaults to high

The effort parameter tunes how hard the model works on a response — an output-level control, not a raw thinking-token budget. It defaults to high across the Claude API and Claude Code, with higher settings (extra/max) for the hardest tasks; on higher effort Claude thinks more frequently and more deeply. Artificial Analysis measured Opus 4.8 at its max effort setting to report the peak 61.4 Intelligence Index score.

Benchmarks vs the competition

BenchmarkOpus 4.8Opus 4.7GPT-5.5Gemini 3.1 Pro
Artificial Analysis Index61.457.360.2
SWE-Bench Verified88.6%87.6%85.1%87.9%
SWE-Bench Pro (agentic)69.2%64.3%58.6%
GPQA Diamond (PhD science)88.4%87.3%86.0%88.2%
AIME 2025 (math)93.6%92.8%95.2%94.0%

Sources: Anthropic Opus 4.8 announcement and Claude API docs, Artificial Analysis Intelligence Index v4.0 (Opus 4.8 measured at max effort). Cross-model figures use each vendor's reported numbers; harnesses differ, so treat small gaps as ties. SWE-Bench Pro is Anthropic-reported (vendor evaluation). · Artificial Analysis

Pricing & access

API

  • Standard: $5 / $25 per 1M tokens
  • Fast Mode: $10 / $50 per 1M tokens (~2.5× speed)
  • Cached input: 90% off
  • Batch: 50% off (24h SLA)

Subscription

  • Claude Pro: $20/mo — limited Opus 4.8
  • Claude Max: $100-200/mo — much higher Opus quota
  • Bedrock / Vertex: same per-token pricing

Opus 4.8 is the priciest closed model in everyday use, but it's flat to Opus 4.7 and undercuts GPT-5.5 ($5/$30) and GPT-5.5 Pro ($30/$180). For privacy or predictable costs, a self-hosted DeepSeek V4 rig trades a one-time hardware spend for unlimited inference.

When to pick Opus 4.8

  • You want the single strongest closed model in June 2026 (#1 Intelligence Index, leads agentic coding).
  • Hard, multi-step agentic work in Claude Code where the 69.2% SWE-Bench Pro edge pays off.
  • Latency-sensitive pipelines that benefit from the optional ~2.5× Fast Mode.
  • Anthropic-stack teams — Opus 4.8 is a drop-in upgrade from Opus 4.7 at the same price.

When to pick something else

FAQ

What is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic's current flagship model, released May 28, 2026 — just 41 days after Opus 4.7. As of June 2026 it is the top model on the Artificial Analysis Intelligence Index (61.4, ahead of GPT-5.5 at 60.2 and Opus 4.7 at 57.3). It scores 88.6% on SWE-Bench Verified and 69.2% on SWE-Bench Pro, keeps Opus 4.7's $5/$25 per-Mtok pricing and 1M-token context, and adds an optional 2.5× Fast Mode and an effort control that defaults to high. It is API-only — accessible via the Anthropic API, Claude apps, AWS Bedrock, and Google Cloud Vertex AI.
Opus 4.8 vs Opus 4.7: what changed?
Opus 4.8 is a drop-in upgrade at the same $5/$25 list price and 1M context as 4.7. SWE-Bench Verified rose from 87.6% to 88.6%, SWE-Bench Pro from 64.3% to 69.2%, and the Artificial Analysis Intelligence Index from 57.3 to 61.4. The new feature is an optional Fast Mode — set it to run roughly 2.5× faster for double the per-token rate ($10/$50 vs $5/$25), and it is about 3× cheaper than Fast Mode on previous Claude models. The effort parameter (high by default, with extra/max for harder work) carries over and remains the recommended way to trade latency for quality. If you are on 4.7 today, moving to 4.8 is a model-ID swap.
How much does Claude Opus 4.8 cost?
Standard Opus 4.8 API pricing is $5 per million input tokens and $25 per million output tokens — flat versus Opus 4.7. The optional Fast Mode is $10 input / $50 output per Mtok (2× the standard rate) in exchange for roughly 2.5× higher throughput. Cached input is heavily discounted (90% off) and the Batch API is 50% off on a 24-hour SLA. Subscription access is via Claude Pro ($20/mo, usage-capped) and Claude Max ($100-200/mo, much higher Opus quota); Bedrock and Vertex charge the same per-token rates.
What is Opus 4.8 Fast Mode?
Fast Mode is an optional setting that runs Opus 4.8 at roughly 2.5× the standard output speed for double the per-token price ($10/$50 instead of $5/$25). Anthropic ships it about 3× cheaper than the equivalent fast mode on previous Claude models. Use it for latency-sensitive agents, interactive coding loops, and high-throughput pipelines where wall-clock time matters more than squeezing the lowest possible per-token cost. For routine work where latency is fine, standard Opus 4.8 is the cheaper default.
What is the effort control?
The effort parameter lets you tune how hard Opus 4.8 works on a response, trading latency for quality. It defaults to high on every surface (Claude API and Claude Code), with higher settings (extra/max) for the hardest tasks — on higher effort Claude thinks more frequently and more deeply. It is an output-level control, separate from a raw thinking-token budget, and is Anthropic's recommended way to configure performance versus speed. Artificial Analysis measured Opus 4.8 at its max effort setting to report peak scores.
Opus 4.8 vs GPT-5.5: which is better?
As of June 2026 Opus 4.8 sits #1 on the Artificial Analysis Intelligence Index at 61.4, narrowly ahead of GPT-5.5 at 60.2. On agentic coding the gap is wider: SWE-Bench Pro 69.2% (Opus 4.8) vs 58.6% (GPT-5.5). Pricing also favors Opus — $5/$25 standard vs GPT-5.5's $5/$30, and far below GPT-5.5 Pro's $30/$180. GPT-5.5 still leads on competition math (AIME) and brings the ChatGPT ecosystem (custom GPTs, plugins). Choose by stack: Anthropic/Claude Code teams pick Opus 4.8, OpenAI/ChatGPT teams pick GPT-5.5.
Can I run Claude Opus 4.8 locally?
No — like all Claude models, Opus 4.8 is API-only and cannot be downloaded or self-hosted. For frontier-class reasoning you can run on your own hardware, the closest open-weight alternatives are DeepSeek V4-Pro (MIT licensed, ~8× H100), Kimi K2.6 (1T MoE, also ~8× H100), and GLM-5 (smaller 4× H100 footprint). None match Opus 4.8's effort control or Intelligence Index, but they land within a handful of points on standard coding benchmarks.

Related models

🎯
AI Learning Path

Go from reading about AI to building with AI

20 structured courses. Hands-on projects. Runs on your machine. Start free.

Or own it for life — Lifetime $149 $599, pay once
LM

Written by the Local AI Master Team

The team behind Local AI Master

We build Local AI Master around practical, testable local AI workflows: model selection, hardware planning, RAG systems, agents, and MLOps. The goal is to turn scattered tutorials into a structured learning path you can follow on your own hardware.

✓ Local AI Curriculum✓ Hands-On Projects✓ Open Source Contributor
More on AI Models for Coding
See the full Best Local AI for Coding guide.
📚
Free · no account required

Grab the AI Starter Kit — career roadmap, cheat sheet, setup guide

No spam. Unsubscribe with one click.

🎯
AI Learning Path

Found your model? Now build something with it.

20 hands-on courses — RAG, agents, fine-tuning — all running locally. First chapter free, no card.

Or own it for life — Lifetime $149 $599, pay once
Free Tools & Calculators