Claude Opus 5
Released
Claude Opus 5 is an AI model released by Anthropic on Friday, Jul 24 2026, 24 days after Claude Sonnet 5. It has a 1M token context window. Benchmark results (shown below) cover Gray Swan IPI, SWE-Bench Verified, SWE-Bench Multilingual, SWE-Bench Multimodal, FrontierCode v1.1 (Main), Next.js Evals, and 19 more.
API pricing
- InputWhat you pay for everything you send the model — your question, plus any documents or earlier conversation you include with it.
- $5.00
- Cached inputA reduced rate for text you send over and over. If every request starts with the same instructions or the same document, the provider keeps a copy ready and charges less to read it again.
- $0.50
- 5 min cache writeA one-off charge for storing text so later requests can reuse it at the cheaper cached rate. This option keeps it for five minutes.
- $6.25
- 1 hr cache writeA one-off charge for storing text so later requests can reuse it at the cheaper cached rate. This option keeps it for an hour, so it costs more than the five-minute one.
- $10.00
- OutputWhat you pay for the text the model writes back. It is normally the dearer half: producing an answer costs more than reading one.
- $25.00
Available from
| Amazon Bedrock | — | $5.00 | $25.00 | 1M |
|---|---|---|---|---|
| Azure | global | $5.00 | $25.00 | 1M |
| Azure | us | $5.00 | $25.00 | 1M |
| global | $5.00 | $25.00 | 1M | |
| Amazon Bedrock | us-east-1 | $5.50 | $27.50 | 1M |
| eu | $5.50 | $27.50 | 1M | |
| us | $5.50 | $27.50 | 1M |
Benchmarks
Coding
SWE-Bench VerifiedCoding — Real coding tasks pulled from open-source projects — the AI has to find and fix actual bugs. A human-checked version of the original SWE-Bench. Higher is better.
96%
via BenchLM
#1Best published SWE-Ver score of all tracked models
SWE-Bench MultilingualMultilingual coding — Like SWE-Bench, but the coding problems span many programming languages, not just one. Tests how broadly the AI can code. Higher is better.
89.5%
#1Best published SWE-Multi score of all tracked models
SWE-Bench MultimodalMultimodal coding — Real bug reports that arrive with pictures attached — a screenshot, a mockup, a page rendering wrongly — so the AI has to read the image as well as the code to work out what to fix. Higher is better.
59.4%
#1Best published SWE-MM score of all tracked models
FrontierCode v1.1 (Main)Agentic coding — A set of very hard, frontier-difficulty coding tasks an AI agent has to complete end to end. The score is the share of tasks in the main split it solves. Higher is better.
53.4%
main split
#1Best published FrontierCode score of all tracked models
Next.js EvalsNext.js coding — Vercel's open eval of how well AI coding agents build and migrate real Next.js apps — measured as the share of tasks the agent completes successfully. Higher is better.
92%
#1Best published Next.js score of all tracked models
Supabase EvalsSupabase coding — Supabase's own open benchmark: a coding agent is dropped into a real Supabase project and asked to do real work — set up a schema, fix a broken security policy, debug an Edge Function — and every run is checked against a live Supabase stack. This is the headline number, where the agent has Supabase's own skills loaded, as most people building on Supabase would. The score is the share of scenarios it got right. Higher is better.
91.3%
with skills
89.9%
no skills
with skills
#1Best published Supabase score of all tracked models
no skills
#1Best published Supabase score of all tracked models
SWE-Bench ProAgentic coding — Can the AI fix real bugs in real software? It's handed actual problems from open-source projects and has to write code that genuinely solves them. Higher is better.
79.2%
#3 of 23Best: Claude Fable 5.1 · 81.2%
DeepSWE 1.1Agentic coding — Artificial Analysis' independent test of deep, agentic software-engineering work — the AI has to plan and carry out substantial coding tasks end to end. (Version 1.1 of the test.) Higher is better.
68.8%
#8 of 25Best: Muse Spark 1.3 · 75.4%
Terminal & CLI
Frontier-Bench v0.1Agentic computer work — A hard, ever-evolving set of real computer tasks — coding, system administration, data work, and more — that an AI agent has to complete on its own. Run by the Harbor / Laude Institute team as the successor to Terminal-Bench (v0.1 is the first release of the task set). The score is the share of tasks solved. Higher is better.
43.3%
#1Best published Frontier v0.1 score of all tracked models
Terminal-Bench 4.0Agentic terminal coding — Can the AI work in a command-line terminal — running commands and finishing technical setup tasks the way a developer would? Version 4.0 recalibrated how much time, CPU and memory each task gets, removed eight tasks and fixed nineteen, so fewer runs fail for reasons that have nothing to do with the model. Scores are not comparable with earlier versions. Higher is better.
51.82%
via Terminal-Bench
#4 of 15Best: Claude Mythos 5.1 · 60.9%
Terminal-Bench-Science 0.1Agentic scientific computing — The same command-line setup as Terminal-Bench, pointed at scientific work: the AI has to drive research tooling and computational workflows through to a result, rather than administer a machine. Version 0.1 is the first release of the task set, and scores run lower than on the general board. Higher is better.
29%
#3 of 4Best: GPT-6 Astra · 64.6%
Agentic & tool use
OSWorld 2.0Agentic computer use — Can the AI actually operate a computer — clicking, typing, and using real apps — to finish tasks on its own? Version 2.0 is a harder, refreshed task set. Higher is better.
70.6%
#1Best published OSWorld 2.0 score of all tracked models
BrowseCompWeb browsing — Can the AI browse the web and track down hard-to-find answers? Higher is better.
90.8%
#4 of 28Best: GPT-5.6 Sol · 92.2%
AutomationBenchBusiness workflows — Tests whether the AI can run real multi-step business workflows — the kind of end-to-end office processes companies want to automate — from start to finish. Higher is better.
26%
#9 of 10Best: Qwen3.8-Max-0902 · 50.8%
Reasoning & science
BioMysteryBenchBiology — Real unsolved-style biology puzzles — the AI has to reason its way to an answer the way a research biologist would. The “hard” split contains the toughest cases. Higher is better.
49.4%
hard
90.1%
human solved
hard
#2 of 3Best: Gemini 3.8 Flash · 56.5%
human solved
#1Best published BioMystery score of all tracked models
Humanity's Last ExamMultidisciplinary reasoning — Humanity's Last Exam — extremely hard expert questions across many subjects, written so you can't just look up the answer. “No tools” means the AI answers on its own. Higher is better.
56.3%
no tools
64.7%
with tools
no tools
#2 of 21Best: Claude Fable 5.1 · 60.9%
with tools
#2 of 41Best: Claude Fable 5.1 · 65%
ARC-AGI-3Novel problem-solving — The third generation of the ARC-AGI series: instead of static puzzles, the AI is dropped into small interactive game-like environments it has never seen and has to figure out the rules and solve them on its own. Higher is better.
30.2%
#2 of 2Best: GPT-6 Astra · 99.9%
Knowledge work
GDPval-AA v2Knowledge work — economically valuable knowledge work (v2, re-based Elo)
1861
#1Best published GDPval v2 score of all tracked models
AA-BriefcaseKnowledge work — Artificial Analysis agentic office-work eval (Elo)
1685
#2 of 4Best: Claude Fable 5.1 · 1694
Legal & tax
Harvey's Legal Agent Benchmark (Held-out)Agentic legal work — Harvey's test of whether an AI agent can complete real legal work, scored on a held-out set of tasks the model makers never see — making the numbers harder to game. Higher is better.
11.7%
Healthcare
HealthBench ProfessionalHealth — Realistic health conversations graded against detailed rubrics written by physicians — can the AI respond the way a careful medical professional would? Higher is better.
59.8%
#3 of 3Best: Claude Fable 5 · 63.3%
Robustness
Gray Swan IPIPrompt injection robustness — Attackers hide malicious instructions inside content the AI reads — a web page, an email, a document — and try to hijack what it does. Gray Swan's indirect prompt injection benchmark measures how often such an attack succeeds when the attacker gets a single try. Lower is better.
0.2%
k = 1
1.6%
k = 10
2%
k = 15
k = 1
#1Best published Gray Swan score of all tracked models
k = 10
#1Best published Gray Swan score of all tracked models
k = 15
#1Best published Gray Swan score of all tracked models
BullshitBench v2Nonsense detection — Given a confidently-worded but nonsensical prompt, does the AI spot that it makes no sense and push back — instead of playing along and inventing an answer? The score is how often it clearly called out the nonsense. Higher is better.
73%
#11 of 74Best: Claude Opus 4.8 · 95%
Community preference
threejsevalCommunity preference (Three.js) — Every model gets the same prompt — "the Eiffel Tower", "a glass fishbowl", "a robot arm picking toys into a box" — and builds a 3D scene in Three.js. Real people then see two scenes side by side, names hidden, and vote for the one they prefer. The votes become a chess-style Elo rating on threejseval.com, averaged across all the prompts. It measures whether the scene looks and moves right to a human eye, not whether the code passes a test. Higher is better.
1770
#3 of 16Best: Claude Fable 5.1 · 2037
Source: Terminal-Bench, retrieved 30 August 2026 · Source: BenchLM, retrieved 24 August 2026. Every other score here is the figure the lab published at launch.
About
Claude Opus 5, released July 24, 2026, brought Anthropic's Claude 5 generation to the Opus tier six weeks after Claude Fable 5 opened it — at half Fable's price, keeping Opus 4.8's $5 per million input tokens and $25 per million output. The pitch was flagship-class agentic capability at workhorse pricing: at launch it scored 43.3% on Frontier-Bench v0.1, more than double Opus 4.8's 21.1% and nearly ten points clear of Fable 5, and posted a GDPval-AA v2 Elo of 1861 for knowledge work, the best published score at the time.
The launch card leaned on breadth: 90.8% on BrowseComp for agentic search, 70.6% on OSWorld 2.0 computer use, 64.7% on Humanity's Last Exam with tools, and 30.2% on ARC-AGI-3 — which Anthropic reported as roughly three times the next best published result on the novel problem-solving benchmark. Anthropic made it the default model on Claude Max and the strongest model available on Claude Pro, positioning Opus 5 as the everyday frontier model while Fable 5 kept the edge on a handful of evaluations, including DeepSWE agentic coding and Harvey's held-out legal benchmark.
Compare Claude Opus 5 with
Suggested comparisons
Claude Opus 5vsClaude Fable 5.1Claude Opus 5vsGPT-6 AstraClaude Opus 5vsGemini 3.8 FlashClaude Opus 5vsMuse Spark 1.3Claude Opus 5vsGrok 4.6Claude Opus 5vsDeepSeek-V4-Pro-0813Claude Opus 5vsMistral Medium 3.5Claude Opus 5vsKimi K3Claude Opus 5vsGLM-5.3-FlashClaude Opus 5vsQwen3.8-Max-0902Claude Opus 5vsNemotron 3.5 Lightning
Frequently asked questions
Claude Opus 5 was released by Anthropic on Friday, Jul 24 2026.