Claude Opus 4.8 Ships 1M Context: Anthropic’s 2026 Coder

Anthropic just shipped Claude Opus 4.8, and the headline number is hard to overstate: a 1-million-token context window on the company’s most capable model, paired with a new Fast mode that delivers Opus-grade output at noticeably higher speed. This is the strongest coding and agentic model Anthropic has released, and the jump from a few hundred thousand tokens to a full million changes which whole-codebase and long-document workflows are actually feasible in one shot. If you build software, run agents, or wrangle large documents for a living, this release resets your assumptions about what fits in a single prompt.

What’s actually new in Claude Opus 4.8

The core of the Anthropic Opus 4.8 release is the Claude Opus 4.8 1M context window. Where earlier Opus generations topped out well below that, 4.8 holds roughly a million tokens in a single request — on the order of a mid-sized codebase, a full quarter of internal docs, or a stack of long PDFs — without manual chunking and stitching. The model ID is claude-opus-4-8, and the extended-context variant is addressed as claude-opus-4-8[1m].

The second piece is Fast mode. Historically the tradeoff was blunt: pick the smartest model and wait, or pick a smaller model and give up quality. Fast mode breaks that. It runs the actual Opus model — not a distilled or downgraded sibling — and serves output faster. In Claude Code you toggle it with /fast, and it’s available on Opus 4.8 and the prior 4.7 release. You keep frontier reasoning while shaving latency on the iterative, back-and-forth work where waiting hurts most.

Opus 4.8 also slots into the broader Claude 5 family. Alongside it sit Sonnet 5 (claude-sonnet-5) as the balanced workhorse, Haiku 4.5 (claude-haiku-4-5-20251001) for high-volume low-latency tasks, and Fable 5 (claude-fable-5) rounding out the lineup. Opus 4.8 is the flagship you reach for when the task is hard, long, or agentic — and the 1M window makes “long” a first-class use case rather than a workaround.

Why the Claude Opus 4.8 1M context matters

• Whole-repo reasoning becomes routine. Load a real service — source, tests, config, and docs — and ask for a cross-cutting refactor without the model losing the thread between files.
• Less glue code, fewer retrieval hacks. Many RAG pipelines exist purely to work around small context. With 1M tokens, “just put it in the prompt” is a legitimate strategy for a large class of problems.
• Long-document workflows stop fragmenting. Contracts, research corpora, and multi-hundred-page manuals fit in one pass, so summaries and extractions reason over the whole thing instead of disconnected slices.
• Agents run longer without amnesia. For Claude Opus 4.8 coding agents, a bigger window means more tool calls, more intermediate state, and more of the plan stays resident before you have to compact.
• Fast mode makes iteration cheap in time. Tight edit-run-review loops get shorter, which compounds across a day of real work.
• Fewer “context-lost” bugs. A meaningful share of agent failures come from truncation and forgotten instructions; a larger window reduces that failure surface directly.

How to use Claude Opus 4.8 today

1. Turn on Fast mode in Claude Code. If you already use the CLI, this is a one-liner in the session:

   /fast

   Run it again to toggle back. It keeps Opus quality and serves output faster — ideal for iterative coding.

2. Call Opus 4.8 from the API. Point your existing Anthropic SDK code at the new model ID:

   curl https://api.anthropic.com/v1/messages \
     -H "x-api-key: $ANTHROPIC_API_KEY" \
     -H "anthropic-version: 2023-06-01" \
     -H "content-type: application/json" \
     -d '{
       "model": "claude-opus-4-8",
       "max_tokens": 2048,
       "messages": [
         {"role": "user", "content": "Refactor this module for testability and explain the tradeoffs."}
       ]
     }'

3. Opt into the 1M context window. To use the extended window, target the 1M variant of the model:

   from anthropic import Anthropic
   
   client = Anthropic()
   
   resp = client.messages.create(
       model="claude-opus-4-8[1m]",
       max_tokens=4096,
       messages=[
           {"role": "user", "content": "Here is our full codebase. Find every place we read this config key and propose a safe migration."}
       ],
   )
   print(resp.content[0].text)

   Check current headers and console settings for how large-context requests are enabled and billed on your account before you push a million tokens in production.

4. Feed it a whole codebase. Concatenate the files you care about (skip node_modules, build artifacts, and lockfiles) and pipe them in as context:

   # collect source into one payload, then send to your script
   find src -type f \( -name "*.ts" -o -name "*.py" \) \
     -not -path "*/node_modules/*" \
     -exec sh -c 'echo "=== $1 ==="; cat "$1"' _ {} \; > context.txt

5. Prompt for whole-repo tasks explicitly. Big context rewards clear instructions. A reliable pattern:

   You have the full repository above. Do three things:
   1. Identify the seam where auth and billing couple.
   2. Propose the minimal refactor to decouple them.
   3. Output a unified diff, touching only files that must change.
   Do not invent files that are not present in the context.

6. Keep costs sane with prompt caching. If you reuse the same large codebase across many queries, cache the static prefix so you aren’t re-paying to process it every call. This is the single biggest lever on Claude Opus 4.8 pricing for repo-scale prompts.

How it compares

Opus 4.8 is the top of the stack, but it isn’t always the right pick. Here’s how it sits against its closest siblings for the Opus 4.8 vs Sonnet 5 decision most teams actually face.

Model	Model ID	Best for	Relative speed	Relative cost
Claude Opus 4.8	claude-opus-4-8	Hardest coding, long-context, agentic work	Fast mode available	Highest tier
Claude Sonnet 5	claude-sonnet-5	Balanced daily driver, most production traffic	Fast	Mid
Claude Haiku 4.5	claude-haiku-4-5-20251001	High-volume, latency-sensitive, cheap tasks	Fastest	Lowest

The rule of thumb: reach for Opus 4.8 when a task is genuinely hard, spans a large codebase, or drives a long agent loop where reasoning quality determines success. Drop to Sonnet 5 for the bulk of everyday requests where it’s more than capable and cheaper, and to Haiku 4.5 when you need speed and volume over depth. Fast mode narrows the old speed gap enough that “use the smart model” is a more defensible default than it used to be — but it doesn’t change the price tier, so cost discipline still matters at scale.

What’s next

The obvious thing to watch is how teams re-architect around a million tokens. Expect a wave of tooling that leans on big context instead of elaborate retrieval — repo-aware agents, long-document assistants, and “just load everything” workflows that were impractical a model generation ago. Some retrieval infrastructure won’t disappear, since you still can’t fit truly massive corpora in a prompt and caching plus targeted retrieval is often cheaper, but the default will shift toward putting more raw material in front of the model.

On the model side, the Claude 5 family gives Anthropic a clear ladder — Fable 5, Sonnet 5, Haiku 4.5, and Opus 4.8 — and capabilities proven at the Opus tier tend to trickle down. Fast mode arriving on both 4.8 and 4.7 hints at how Anthropic treats speed as a toggle rather than a separate model, and that framing will likely spread across the lineup.

The practical watch items for builders: how large-context pricing and caching evolve, how far agent frameworks push the 1M window in real deployments, and whether Fast mode changes which model teams default to. If you run anything agentic or codebase-scale, benchmark Opus 4.8 against your current setup now and see where the bigger window removes complexity you’ve been carrying.

Frequently Asked Questions

What is Claude Opus 4.8?

Claude Opus 4.8 is Anthropic’s flagship model in the Claude 5 family and its strongest release for coding and agentic tasks. It ships with a 1-million-token context window and a Fast mode that delivers Opus-level quality at higher output speed. Its model ID is claude-opus-4-8.

How big is the Claude Opus 4.8 1M context window really?

One million tokens is roughly on the order of a mid-sized codebase or several long documents at once. In practice you can load whole repositories, large document sets, or extended agent histories into a single request instead of chunking them and reassembling the results.

What is Fast mode and does it lower quality?

Fast mode serves the actual Opus model with faster output — it does not swap in a smaller or downgraded model. You keep frontier reasoning and simply wait less. In Claude Code you toggle it with /fast, and it’s available on Opus 4.8 and 4.7.

Opus 4.8 vs Sonnet 5 — which should I use?

Use Opus 4.8 for the hardest problems: complex coding, long-context reasoning, and agent loops where quality is decisive. Use Sonnet 5 (claude-sonnet-5) for most everyday production traffic, where it’s highly capable and more cost-effective. Match the model to task difficulty rather than defaulting to the flagship for everything.

How do I access the 1M context version?

Target the extended-context variant, addressed as claude-opus-4-8[1m], and confirm how large-context requests are enabled and billed for your account in the console. For repeated queries over the same large payload, use prompt caching so you aren’t reprocessing the full context on every call.

What about Claude Opus 4.8 pricing?

Opus 4.8 sits in Anthropic’s top pricing tier, above Sonnet 5 and Haiku 4.5. Because exact rates and large-context billing can change, check your Anthropic console for current numbers before committing at scale — and lean on prompt caching to keep repo-scale prompts economical.