Claude Opus 5: Anthropic's Frontier Model at Half the Price of Fable 5

Anthropic shipped Claude Opus 5 on July 24, 2026, and the headline isn't a new benchmark record — it's the price tag. Opus 5 costs exactly what its predecessor, Opus 4.8, cost ($5 per million input tokens, $25 per million output tokens), yet on several coding and reasoning evaluations it beats Anthropic's own bigger, pricier flagship, Claude Fable 5, which runs roughly twice as expensive per token. That combination — frontier-class results at last-generation pricing — is the actual news here.

If you build on the Claude API, run agentic coding workflows, or are simply deciding which model deserves your budget this quarter, Opus 5 changes the math. Here's what actually shipped, how it stacks up against Fable 5 and OpenAI's GPT-5.6 Sol, and where the real catch is.

Quick Take

• What it is: Anthropic's fourth model in the Claude 5 family, replacing Opus 4.8, with a 1-million-token context window (the only size available — there's no smaller variant) and 128K max output.
• Pricing: $5 per million input tokens, $25 per million output tokens — unchanged from Opus 4.8, and about half of Claude Fable 5's usage-credit rate.
• Why it matters: On Anthropic's own Frontier-Bench and GDPval-AA evaluations, Opus 5 is now the company's state-of-the-art model for coding and knowledge work, ahead of Fable 5 on several tasks despite costing less.
• The catch: Opus 5 still trails Anthropic's Mythos 5 on cybersecurity tasks, and its new "fast mode" (up to 2.5x faster) is Claude API-only — it isn't available yet on Bedrock, Google Cloud, or Microsoft Foundry.

What Anthropic Actually Shipped

Opus 5 has "thinking" turned on by default — a change from Opus 4.8, where extended reasoning had to be explicitly requested. According to Anthropic's official announcement, the model's biggest gains are in agentic coding and long-horizon tasks: staying on task through extended tool-use loops, finishing multi-file features and larger refactors without leaving stubs behind, and catching real bugs during code review at a low false-positive rate. Anthropic also highlights stronger multi-agent coordination, meaning teams of subagents overwrite each other's work less often — a real pain point in agentic frameworks.

It's available immediately through the Claude API, Claude.ai (Pro, Max, Team, and Enterprise plans), Claude Code, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Opus 4.8 stays available everywhere too, so nothing already built on it breaks.

The Pricing Story, in One Table

The number that matters most: Opus 5 didn't get more expensive to get better.

Model	Input / Output (per 1M tokens)	Context window
Claude Opus 5	$5 / $25	1M tokens
Claude Fable 5	$10 / $50 (usage credits)	Larger, but pricier per task
GPT-5.6 Sol	$5 / $30	~1.05M tokens

Pricing verified July 2026 against Anthropic's and OpenAI's published rates; confirm current numbers directly with each vendor before budgeting, since Fable 5's terms have shifted more than once this month.

Fable 5's pricing is worth a caveat: it briefly went dark in June under US export controls, returned July 1, then moved most non-subscriber usage to prepaid credits on July 12 at $10/$50 per million tokens. Opus 5 sidesteps that volatility entirely — same price as the model it replaces, available everywhere Claude already ships.

How It Actually Benchmarks

Independent and Anthropic-reported scores put Opus 5 ahead of both Fable 5 and GPT-5.6 Sol on agentic coding specifically, though not across the board:

Benchmark	Claude Opus 5	Claude Fable 5	GPT-5.6 Sol
Frontier-Bench v0.1 (agentic coding)	43.3%	33.7%	34.4%
OSWorld 2.0 (computer use)	70.6%	66.1%	—
ARC-AGI-3	30.2% (highest reported)	—	—

Opus 5 also converts extra "effort" into meaningfully better results — Anthropic says this is truer of Opus 5 than any earlier Opus model, which is exactly why the effort setting deserves its own section below. The honest caveat, stated by Anthropic itself: Opus 5 remains behind Mythos 5, Anthropic's specialized model, on cybersecurity-specific tasks. Nobody's model wins everything, and it's a good sign that Anthropic says so directly rather than only publishing favorable comparisons.

The Effort Toggle Is the Real Story for Cost Control

Opus 5 ships with five effort levels — low, medium, high, xhigh, and max — defaulting to high. This is the lever that actually controls your bill: Anthropic reports that low and medium effort produce strong quality at a fraction of the tokens and latency of higher settings, while xhigh and max exist for the hardest reasoning and agentic work. There's a real trade-off buried in here, too — disabling extended thinking is only allowed at high effort or below; ask for xhigh or max with thinking disabled and the API returns an error.

Anthropic also introduced a "fast mode" for Opus 5, priced at $10/$25 more per million tokens ($10 input / $50 output total) for roughly 2.5x the speed. It's a research preview available only through the Claude API for now — teams on Bedrock, Google Cloud, or Microsoft Foundry don't get it yet, which is a real gap if your infrastructure already lives on one of those platforms.

Timing: This Launch Has an IPO Behind It

Anthropic filed confidential paperwork for a public offering in June 2026, with OpenAI reportedly doing the same shortly after. Shipping a model that undercuts your own larger flagship on cost while topping your internal benchmarks is a straightforward way to keep the "AI race" narrative in your favor heading into that process — and it lands the same week Anthropic's biggest rival is managing a security disclosure of its own. None of that changes whether Opus 5 is good for your workload, but it explains why Anthropic chose now to ship a cost story instead of just a capability one.

Should You Actually Switch?

If you're running agentic coding workloads through Claude Code or building on the Claude API, Opus 5 is close to a straightforward upgrade — same price as Opus 4.8, better benchmarks, and it becomes the default model for Claude Max users automatically. If you were paying Fable 5's usage-credit rate for tasks that don't need its specific edge, Opus 5's Frontier-Bench and OSWorld scores suggest you can drop down a tier and save money without losing much.

If cybersecurity-specific work is your priority, Anthropic's own numbers say stick with Mythos 5. And if you're comparing across vendors rather than within Anthropic's lineup, see our full breakdown of Claude vs. ChatGPT vs. Gemini — Opus 5's Frontier-Bench lead over GPT-5.6 Sol is real, but it's one benchmark, not a verdict on every task. Budget-conscious teams comparing open alternatives should also note that Moonshot's Kimi K3 finally shipped its full open weights this week, expanding the field further.

Frequently Asked Questions

Is Claude Opus 5 more expensive than Opus 4.8? No. Pricing is unchanged at $5 per million input tokens and $25 per million output tokens. The new "fast mode" costs extra ($10/$50 per million tokens) but is optional.

Is Claude Opus 5 better than Claude Fable 5? On agentic coding (Frontier-Bench v0.1) and computer-use tasks (OSWorld 2.0), Opus 5 scores higher than Fable 5 while costing about half as much per token. Fable 5 remains Anthropic's larger model overall; Anthropic itself frames Opus 5 as reaching "frontier intelligence at half the cost," not as a strictly better model in every category.

How does Opus 5 compare to GPT-5.6 Sol? Opus 5 leads GPT-5.6 Sol on Frontier-Bench v0.1 (43.3% vs. 34.4%) and is priced slightly lower on output tokens ($25 vs. $30 per million). Input pricing is identical at $5 per million tokens.

Where can I use Claude Opus 5? The Claude API, Claude.ai (Pro, Max, Team, Enterprise), Claude Code, Amazon Bedrock, Google Cloud, and Microsoft Foundry, effective July 24, 2026.

What is the effort toggle and why should I care? It's a five-level dial (low to max) that trades reasoning depth for token cost and speed. Anthropic says lower effort levels now hold up better on Opus 5 than on prior models, making it a genuine cost-control tool rather than just a quality knob.

Bottom Line

Claude Opus 5 is a rare case of a model getting measurably better without getting more expensive — it beats Fable 5 on several coding and computer-use benchmarks at roughly half the per-token cost, and it undercuts GPT-5.6 Sol on output pricing while leading it on agentic coding. The real limitations are narrow: it's still behind Mythos 5 on cybersecurity work, and fast mode hasn't reached Bedrock, Google Cloud, or Microsoft Foundry yet. For anyone already spending on Claude, this is a low-risk upgrade worth testing this week. For a wider view of how the major assistants compare, browse our AI Tools coverage or see 10 AI tools actually worth paying for in 2026.