Technology RadarTechnology Radar

Claude Opus 5

frontierllmcodingagenticreasoningmultimodal
Adopt

Claude Opus 5, released July 24, 2026, is the new default Opus model — exceeding Fable 5 on key coding benchmarks while matching Opus 4.8's price point at $5/$25 per MTok. It supersedes Opus 4.8 for most agentic workloads.

Why It's in Adopt

Opus 5 lands immediately in Adopt for the same reasons Opus 4.8 did — but with a stronger benchmark profile and the same price:

  • Frontier-Bench v0.1: 43.3% — beats Fable 5 (33.7%) on this coding-focused benchmark; more than doubles Opus 4.8's score
  • ARC-AGI 3: 30.2% — 20× improvement over Opus 4.8 on novel problem-solving
  • CursorBench 3.2: within 0.5% of Fable 5 at max effort — effectively peer-level with the flagship on this agentic coding eval
  • Self-verification: checks its own work iteratively and catches errors before surfacing results — the most practically meaningful behavioral upgrade over prior Opus versions
  • 1M context window — full parity with Fable 5 and Opus 4.8
  • Default Opus model in Claude Code (v2.1.219, July 24, 2026) — confirmed callable via claude-opus-5 API ID; /fast applies to Opus 5 at $10/$50 per MTok

At the same base price as Opus 4.8 with meaningfully stronger benchmarks and self-verification behavior, there is no practical reason to prefer Opus 4.8 for new deployments.

Performance

Benchmark Opus 5 Opus 4.8 Fable 5
Frontier-Bench v0.1 43.3% ~18% 33.7%
ARC-AGI 3 30.2% ~1.5% —
CursorBench 3.2 Within 0.5% of Fable 5 — (baseline)

Note: Anthropic has shifted toward Frontier-Bench v0.1 and ARC-AGI 3 as primary benchmarks for Opus 5. Secondary benchmarking sites (llm-stats.com, morphllm.com) report 97% SWE-bench Verified, which would place Opus 5 above Fable 5 (95%) — these are not yet independently verified against a primary Anthropic source.

Pricing

Property Value
Input $5 per million tokens
Output $25 per million tokens
Fast mode input $10 per million tokens
Fast mode output $50 per million tokens
Context window 1,000,000 tokens
Prompt caching Up to 90% savings on repeated context
Batch processing 50% savings for async tasks

Opus 5 and Opus 4.8 share identical base pricing. Fast mode (research preview) is available on both at $10/$50 per MTok. Fast mode is not available on Opus 4.7 (returns an error).

How to Access

  • Claude.ai — available on Pro and higher plans
  • Anthropic API — claude-opus-5 model ID
  • Claude Code — default Opus model as of v2.1.219 (July 24, 2026); accessible via alias "opus" or explicit claude-opus-5
  • Amazon Bedrock — check availability
  • Google Cloud Vertex AI — check availability

Key Characteristics

Property Value
License Proprietary
Context window 1,000,000 tokens (1M)
Pricing $5/M input, $25/M output
Fast mode pricing $10/M input, $50/M output
Provider Anthropic
API model ID claude-opus-5
Release date July 24, 2026
Tokenizer Newer (~30% more tokens than pre-4.7 models)

Further Reading