Anthropic

Claude Opus 4.8

Type
language model
Context
1,000K tokens
Max output
128K
Released
28 May 2026
Knowledge cutoff
December 2025
API string
anthropic/claude-opus-4-8

Prices by provider

Provider Input, $ per 1M Output, $ per 1M In our data since
unorouter $0.425 $2.125 31 Jul 2026
xpersona $1.5 $9.25 31 Jul 2026
Poe $4.293 $21.465 31 Jul 2026
NanoGPT $4.998 $25.007 31 Jul 2026
daoxe $5 $25 31 Jul 2026
neon $5 $25 31 Jul 2026
google-vertex-anthropic $5 $25 31 Jul 2026
ofox $5 $25 31 Jul 2026
routing-run $5 $25 31 Jul 2026
freemodel $5 $25 31 Jul 2026
modelis $5 $25 2 Aug 2026
AIHubMix $5 $25 31 Jul 2026
Abacus.AI $5 $25 31 Jul 2026
Amazon Bedrock anthropic.claude-opus-4-8 $5 $25 31 Jul 2026
Anthropic $5 $25 31 Jul 2026
Azure Cognitive Services $5 $25 31 Jul 2026
Cloudflare AI Gateway $5 $25 31 Jul 2026
CrossModel $5 $25 31 Jul 2026
DigitalOcean Gradient AI $5 $25 1 Aug 2026
FastRouter $5 $25 31 Jul 2026
GMI Cloud $5 $25 31 Jul 2026
GitHub Copilot $5 $25 31 Jul 2026
Google Vertex AI $5 $25 31 Jul 2026
LLM Gateway $5 $25 31 Jul 2026
Merge Gateway $5 $25 31 Jul 2026
Microsoft Foundry $5 $25 31 Jul 2026
OpenCode Zen $5 $25 31 Jul 2026
OpenRouter $5 $25 31 Jul 2026
Pioneer $5 $25 31 Jul 2026
SAP AI Core $5 $25 1 Aug 2026
Vercel $5 $25 31 Jul 2026
ZenMux $5 $25 31 Jul 2026
Amazon Bedrock eu.anthropic.claude-opus-4-8 $5.5 $27.5 31 Jul 2026
Venice AI $6 $30 31 Jul 2026

Only the base tier and only per-token prices are shown. Batch, discounted and cached rates, as well as prices per image or per second of video, do not go into this table: they cannot stand in the same column as a price per million tokens.

A single provider appears several times if it sells this model under different identifiers and at different prices — for example, at different weight precision. The identifier is shown next to the name. Identical offers that arrived from two sources under different spellings are merged into one row.

The date in the last column is the day this price first entered our collection. The price may well be older: before that day we simply were not recording it. It has not changed since — otherwise a new row with a new date would stand in its place.

Measurement results

Task set Result Run conditions Measured by
Arena Score, web development 1,538.84 1,531–1,547
Arena Score, expert questions 1,489.04 1,478–1,500
Arena Score, programming 1,485.6 1,478–1,493
Arena Score, long queries 1,480.54 1,474–1,487
Arena Score, multi-turn dialogue 1,475.63 1,467–1,484
Arena Score, hard prompts 1,473.3 1,467–1,479
Arena Score in French 1,471.76 1,453–1,490
Arena Score, mathematics 1,466.9 1,452–1,482
Arena Score in Russian 1,464.49 1,454–1,475
Arena Score, instruction following 1,461.02 1,454–1,468
Arena Score in English 1,457.59 1,451–1,464
Arena Score, overall 1,453.02 1,448–1,458
Arena Score in Spanish 1,449.21 1,429–1,469
Arena Score, creative writing 1,443.1 1,434–1,452
Arena Score, understanding diagrams 1,304.65 1,293–1,317
Arena Score, text recognition in images 1,297.38 1,289–1,306
Arena Score, working with images 1,287.87 1,280–1,296
Mock AIME 2024–2025 — olympiad problems 98.33 % effort: max · with a tuned harness Epoch evaluations
ARC-AGI — generalising to unseen patterns 92.5 % effort: max · with a tuned harness https://arcprize.org/leaderboard
GPQA Diamond — graduate-level questions 88.05 % effort: max · with a tuned harness Epoch evaluations
WeirdML — unusual machine learning tasks 82.89 % effort: xhigh https://htihle.github.io/weirdml.html
FrontierMath, levels 1–3 80 % effort: max Epoch evaluations
ARC-AGI-2 72.08 % effort: high
CursorBench — edits in the editor 63.8 % effort: max https://cursor.com/cursorbench
SimpleBench — trick questions 57.76 % https://lmcouncil.ai/benchmarks self-reported
FrontierMath, level 4 — research-grade problems 56.1 % effort: max Epoch evaluations
DeepResearch Bench — deep research 50.23 % effort: high https://drb.futuresearch.ai/#drb self-reported
GSO-Bench — code optimisation 47.06 % https://gso-bench.github.io/index.html
FrontierCode — patches fit to be merged into a project 46.5 % https://cognition.com/frontiercode
APEX-Agents 42.5 % effort: max
SimpleQA Verified — factual accuracy 39.5 % effort: max Epoch evaluations
Chess puzzles 30.56 % effort: max Epoch evaluations
CritPt — physics problems 20.86 % effort: max
OSWorld 2.0 — working inside an operating system 20.6 % effort: max https://osworld-v2.xlang.ai/ self-reported
Remote Labor Index — jobs from a freelance marketplace 8.33 %
Arena Score, agent tasks 0.03 0–0