Anthropic Launches Claude Opus 5: Near-Fable Intelligence at Half the Price

TLDR

Anthropic on July 24, 2026 launched Claude Opus 5, calling it a step-change for the Opus tier that comes close to Claude Fable 5 frontier intelligence at half the price. Opus 5 is the new default on Claude Max, the strongest model on Claude Pro, and is available everywhere Claude runs at $5 / $25 per million input/output tokens (same as Opus 4.8), with optional Fast mode (~2.5× speed at 2× base price).

Product and pricing

Item	Detail
Model ID	claude-opus-5
Price	$5 / $25 per MTok input/output (unchanged from Opus 4.8)
Fast mode	~2.5× default speed; 2× base price on Claude Platform / usage credits in Claude Code
Plan placement	Default on Max; strongest model on Pro
API extras (beta)	Mid-conversation tool changes without busting prompt cache; automatic fallbacks when safety classifiers flag Opus 5 / Fable 5

Anthropic frames Opus 5 for everyday professional use: long-running agents, coding, and knowledge work, with stronger self-verification and iteration than Opus 4.8. Early-access quotes from Cursor, Devin, Zapier, Lovable, Box, and others emphasize debugging, end-to-end automation, and lower variance.

Lab-reported performance (primary)

From Anthropic’s launch post:

• Frontier-Bench v0.1 and GDPval-AA: Opus 5 claimed new SOTA on coding/knowledge-work style boards (still behind Mythos 5 on cybersecurity).
• CursorBench 3.2 (max effort): within ~0.5% of Fable 5 peak at half the cost per task.
• ARC-AGI 3: ~3× next-best model (lab chart).
• OSWorld 2.0: best cost-performance; lab says it can beat Fable 5’s best result at just over one-third the cost.
• Life sciences: higher than Opus 4.8 on internal organic chemistry / protein-function style tasks.
• Alignment: automated behavioral audit scores Opus 5 as Anthropic’s most aligned recent model (lowest overall misaligned behavior in their chart).

Safeguards: cyber classifiers are less restrictive than Fable 5 (~85% fewer interventions expected) but still block binary-based vuln scanning, pen-testing, and exploit generation; CVP enterprises get a less-restricted path. Biology blocked on Fable now routes to Opus 5 rather than Opus 4.8. Opus 5 does not advance the dual-use cyber/bio frontier relative to Mythos 5 (system card).

Independent rankings

Source	Snapshot	Standing
Artificial Analysis Intelligence Index	Public mirror cited July 24, 2026 (BenchLM display)	Claude Opus 5 ~60.7% — listed ahead of Claude Fable 5 (~59.9%) and GPT-5.6 Sol (~58.9%) among ~167 models on that snapshot
LLM Stats / aggregate boards	Day-of listings	Opus 5 appears as a new high-rank proprietary entry; full Arena Elo still maturing right after launch

Note: Day-one independent boards can move quickly; treat AA index figures as a dated snapshot, not a permanent ranking. Lab CursorBench / Frontier-Bench claims should be read beside third-party harnesses as votes accumulate on LMArena and coding arenas.

Why this story matters

Opus 5 is Anthropic’s bid to make “near-Fable” the daily default—not a scarce frontier SKU. If independent arenas confirm cost-performance close to Fable at Opus price, Max/Pro defaults and agent platforms (Cursor, Devin, Claude Code) reprice the mid-frontier stack against GPT-5.6 Sol and open-weight challengers. Watch fallbacks under cyber classifiers, Fast mode economics, and whether AA / Arena hold Opus 5 above Sol after the launch week noise settles.

Sources

• Anthropic: Introducing Claude Opus 5 (July 24, 2026)
• Anthropic: Claude Opus 5 System Card
• Artificial Analysis / BenchLM mirror of Intelligence Index snapshot citing Opus 5 (July 24, 2026)
• Coursiv: Claude Opus 5 benchmarks & pricing guide (updated July 25, 2026)

Featured Image Alt Text

Engraved Claude geometric mark as a fifth-generation Opus seal with agent-workflow lines and a half-price scale motif versus a Fable crest.

Tags

Anthropic, Claude, Opus 5, Fable 5, Models, Coding, Agents, Artificial Analysis, July 24