Modelpeers
Claude Opus 4.8 charactervsClaude Opus 5 character

Claude Opus 4.8 vs Claude Opus 5: Should you upgrade?

Facts checked Oct 22, 2026. Every number links to its source.

Summary

Derived from sourced facts, not from builder opinions
  • Opus 5 costs the same as Opus 4.8: $5 input and $25 output per million tokens.
  • With Claude Code as the agent, Opus 5 scored 53.94% on Terminal-Bench 4.0. Opus 4.8 scored 23.64%.
  • One breaking change: on Opus 5, turning thinking off at xhigh or max effort returns a 400 error.
  • Both are now legacy models. Anthropic suggests moving to Opus 5.5, which costs less than either.

Choose Claude Opus 4.8 if

  • Your app turns thinking off at xhigh or max effort, and you cannot change that yet.
  • You have tuned prompts on Opus 4.8 and cannot retest before mid-2027.

Choose Claude Opus 5 if

  • You want much higher coding scores at the same price.
  • You want the newer knowledge cutoff (May 2026 against January 2026).
  • You want a later retirement date (no sooner than July 24, 2027).
Claude Opus 4.8
Claude Opus 5

Follow both to get an email when either one changes its price, limits or behavior.

Best by task

Agentic coding (Terminal-Bench 4.0 official board, both in Claude Code)

Claude Opus 5

53.94% at xhigh against 23.64% at max, 330 trials each.

Agentic coding (Terminal-Bench 4.0, Vals AI, Mini-SWE-agent harness)

Claude Opus 5

53.53% against 23.23%, both models run through the API in the same harness.

SWE-bench Verified (Vals AI, Mini-SWE-agent harness, archived)

Claude Opus 5

97.00% against 88.60% before Vals AI archived the benchmark.

Overall capability (Epoch Capabilities Index)

Tied

163.94 against 159.01. Opus 5 is ahead, but the 95% confidence intervals overlap slightly (161.12 to 168.24 and 156.89 to 161.95).

Cost per solved task on Terminal-Bench 4.0 (Vals AI, API prices)

Claude Opus 5

Opus 5 cost $18.60 per test and Opus 4.8 $17.14, but Opus 5 solved 53.53% of tasks against 23.23%. Per solved task that is about $35 against $74, derived from Vals AI's figures.

Cost

One request in an agent loop that rereads a repository, so most input is cached. A full task makes many such requests. Illustrative, not measured.

Claude Opus 4.8

$0.170 per request

$170.00 a month for 1,000 requests

Claude Opus 5

$0.170 per request

$170.00 a month for 1,000 requests

Estimate per request: (input × input rate + cached input × cached rate + output × output rate) ÷ 1,000,000, at list prices on the cited pricing pages. It is not a measured task cost: models use different numbers of tokens, including reasoning tokens, for the same task.

Side by side

Claude Opus 4.8Claude Opus 5
MakerAnthropic
Anthropic
Model IDclaude-opus-4-8 (Bedrock: anthropic.claude-opus-4-8)
claude-opus-5 (Bedrock: anthropic.claude-opus-5)
Released2026-05-28
2026-07-24
StatusActive (legacy). Anthropic suggests moving to Opus 5.5
Active (legacy)
API price per 1M tokens$5 input / $25 output · 5-minute cache write $6.25, 1-hour cache write $10, cache read $0.50 · Batch 50% off
$5 input / $25 output · 5-minute cache write $6.25, 1-hour cache write $10, cache read $0.50 · Batch 50% off
Fast mode price$10 input / $50 output (research preview on the API)
Not published
Context window1M tokens
1M tokens
Max output128K tokens (300K on Batch API with beta header)
128K tokens (300K on Batch API with beta header)
Knowledge cutoffJanuary 2026 (reliable knowledge and training data)
May 2026 (reliable knowledge and training data)
Reasoning settingsAdaptive thinking, default effort high
high · adaptive thinking (can be turned off, unlike Opus 5.5)
Where to use itClaude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS
Claude API, Amazon Bedrock, Google Cloud (Google managed and Anthropic managed), Microsoft Foundry, Claude Platform on AWS
Retires no sooner than2027-05-28
2027-07-24
Terminal-Bench 4.0 (official board)23.64%Claude Code, max effort, 330 trials53.94%at xhigh (rank 11), 51.82% at max, Claude Code, 330 trials
Terminal-Bench 4.0 (Vals AI, same harness for all)23.23%rank 26, $17.14 per test53.53%rank 8, $18.60 per test
SWE-bench Verified88.60%with Mini-SWE-agent, rank 12 on the archived Vals AI board (as of 2026-09-01)97.00%with Mini-SWE-agent, rank 1 on the archived Vals AI board (as of 2026-09-01)
Epoch Capabilities Index159.01(CI 156.89 to 161.95)163.94(CI 161.12 to 168.24)

Scores appear side by side only when both were published for the same test. Settings differ between sources, so small gaps can reverse under another harness.

Updates

Both Claude Opus 4.8 and Claude Opus 5

Retiring

Copilot to remove Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra on November 16

Copilot will deprecate Claude Opus 4.8 and Claude Opus 5 (replace with Claude Opus 5.5) and GPT-5.6 Terra (replace with GPT-6.1 Sol or GPT-6 Luna) on November 16, 2026. Opus 4.8 stays for individual annual-plan subscribers.

Takes effect Nov 16, 2026 · in 15 days

Who is affected:
Copilot users and integrations pinned to those models.
Action needed:
Move to the suggested alternatives before November 16, 2026.

Claude Opus 4.8

2 recorded incidents in the last 90 days.

Retiring

Claude Opus 4.8 fast mode is deprecated and ends November 30, 2026

Fast mode for Claude Opus 4.8 is deprecated and will be removed on November 30, 2026. After removal, requests with speed set to fast will return an error.

Takes effect Nov 30, 2026 · in 29 days

Who is affected:
Developers who send fast mode requests to Claude Opus 4.8 will get errors after removal.
Action needed:
Migrate fast mode requests to Claude Opus 5.5 before November 30, 2026.

Claude Opus 5

5 recorded incidents in the last 90 days.

Switching checklist

  1. Change the model ID from claude-opus-4-8 to claude-opus-5 (on Bedrock, anthropic.claude-opus-5).
  2. Search your code for thinking set to disabled together with effort xhigh or max. That returns a 400 error on Opus 5.
  3. Check prompts that depend on recent events: Opus 5 knows events through May 2026, Opus 4.8 through January 2026.
  4. Keep your budget: both cost $5 input and $25 output per million tokens, with the same cache and batch discounts.
  5. Rerun your evaluation set and compare cost per finished task, not cost per token.
  6. Consider going straight to Opus 5.5. From Opus 4.8 its migration guide adds that thinking cannot be turned off, Priority Tier is not supported and max_tokens should be at least 64k at xhigh or max effort.

FAQ

Is Claude Opus 5 more expensive than Opus 4.8?

No. Both cost $5 input and $25 output per million tokens.

Is Opus 5 better at coding than Opus 4.8?

On Terminal-Bench 4.0 in Claude Code, Opus 5 scored 53.94% and Opus 4.8 scored 23.64%.

When will Claude Opus 4.8 be retired?

No retirement is announced. Anthropic lists Opus 4.8 as legacy, retiring no sooner than May 28, 2027, with at least 60 days' notice.

Should I skip Opus 5 and move to Opus 5.5?

Anthropic's Opus 4.8 page suggests migrating to Opus 5.5, which costs $4 and $20 per million tokens.

Which has the newer knowledge cutoff?

Opus 5: May 2026, against January 2026 for Opus 4.8.

Do Opus 4.8 and Opus 5 have the same context window?

Yes. Both take 1M tokens with 128K of output, or 300K on the Batch API with a beta header.

Which costs less per test?

Opus 4.8, at $17.14 per test against $18.60 for Opus 5 in Vals AI's Terminal-Bench 4.0 run, where Opus 5 scored more than twice as high.

Related comparisons

Claude Opus 5 vs Claude Opus 5.5Claude Opus 4.8 pageClaude Opus 5 page

Sources and method

Prices, specs and limits come from each maker's own pages. Benchmark scores come from independent benchmark publishers and the makers' announcements, with the setting each was measured in. Changes come from changelogs, deprecation pages and status pages. The summary is written from these facts only.