Claude Opus 5 deep dive: Anthropic’s reasoning-and-coding flagship at roughly half of Fable

What it is

Claude Opus 5 is Anthropic’s flagship SKU in the AI Hippo catalog for demanding reasoning, coding, and long-horizon agentic work. Catalog copy highlights end-to-end software tasks, code review and bug finding, and visual analysis. Specs: a 1M-token context window; text, image, and file inputs with text output; agent-friendly parameters including reasoning, reasoning_effort, tools, and structured_outputs.

Site `ops_model_first_seen` records Opus 5 and Opus 5 Fast on 2026-07-26—newer than Sonnet 5 and than our existing Fable 5 piece. Start by separating Anthropic’s three lines: Sonnet (throughput/cost), Opus (flagship balance), and Fable/Mythos-class (top price tier).

Spec sheet at a glance

By the numbers (`core_model`): context length 1,000,000; modality `text+image+file->text`. Regular Opus 5 exposes 12 API parameters: include_reasoning, max_completion_tokens, max_tokens, reasoning, reasoning_effort, response_format, stop, structured_outputs, temperature, tool_choice, tools, and verbosity.

The Fast sibling (`anthropic/claude-opus-5-fast`) shares the 1M window and same modalities, with a slightly narrower parameter list (no temperature, no max_completion_tokens). Catalog copy states identical capabilities, higher output speed, and 2× pricing versus regular Opus 5. Knowledge cutoff is undisclosed in this snapshot; verified hosts show no separate batch / cached-input tier (anthropic-direct cached field is 0). Vendor: Anthropic.

Pricing and where to run it

Regular Opus 5: $5 input and $25 output per 1M tokens. This snapshot matches on the OpenRouter primary listing (is_primary_listing=1) and anthropic-direct (model_bound_verified). Output costs 5× input—long generations and agent loops inflate the bill.

Opus 5 Fast: OpenRouter primary listing at $10/$50—exactly 2× regular—same sticker tier as Fable 5. Treat Fast as a same-capability throughput premium, not a different model family.

This snapshot does not yet verify Opus 5 rows on Bedrock/DeepInfra (Sonnet 5 and Fable 5 do have multi-host coverage). Re-check host coverage on the Token page before production.

Inside the family: Sonnet 5 → Opus 5 → Fable 5

Anthropic ladder in this snapshot (OpenRouter primary listing, 1M context aligned): Sonnet 5 at $2/$10 (blend ~$6), Opus 5 at $5/$25 (blend ~$15), Fable 5 and Opus 5 Fast both at $10/$50 (blend ~$30). Versus Sonnet, Opus is about 2.5× on both input and output; versus Opus, Fable/Fast is another 2×.

All three share 1M context, multimodal I/O, and reasoning/tools; regular Opus additionally exposes temperature. On AI Hippo’s composite board (a context-per-dollar proxy, not a capability benchmark): Sonnet 5 sits around #21; Opus 5 and Fable 5 are still outside this snapshot’s unified Top—high unit price depresses that proxy score.

Rule of thumb: default throughput and budget → Sonnet 5; catalog-flagship reasoning/coding at ~2.5× Sonnet → Opus 5; top sticker tier or “same sticker, buy speed” → Fable 5 / Opus Fast.

How it compares across vendors

Blended cost (OpenRouter primary listing mean): Opus 5 about $15 per 1M. Peers: GPT-5.6 Sol about $17.50 (1.05M ctx), GPT-5.4 about $8.75 (1.05M), Kimi K3 about $9 (1.0M), Gemini 3.1 Pro Preview about $7 (1.0M), Grok 4.5 about $4 (0.5M), Llama 4 Scout about $0.20 (1.3M).

Opus lands in “pricier than mainstream multimodal, still clearly under the Fable/Sol sticker ceiling.” It is outside this snapshot’s composite Top—again: that score rewards cheap long context, not arena capability. Judge Opus on flagship/agent/coding fit, not proxy rank. These are spec and cost comparisons, not capability benchmarks.

Cost in practice: two scenarios

Modeled with this snapshot’s OpenRouter primary listing (excluding extra reasoning tokens).

Scenario A—agent coding turn (100K input + 20K output): Opus 5 about $1.00 ($0.50+$0.50); Sonnet 5 about $0.40; Kimi K3 about $0.60; GPT-5.4 about $0.55; GPT-5.6 Sol about $1.10; Fable 5 / Opus Fast about $2.00; Grok 4.5 about $0.32.

Scenario B—long-document pass (500K input + 5K output): Opus 5 about $2.62; Sonnet about $1.05; K3 about $1.57; Sol about $2.65; Fable/Fast about $5.25; GPT-5.4 about $1.32.

Takeaway: Opus sits near Sol, roughly half of Fable, and about 2.5× Sonnet. The second 2× jump from Opus to Fast/Fable hurts most on output-heavy loops. These are spec and cost comparisons, not capability benchmarks.

When to choose it—and when not to

Choose Opus 5 when you want Anthropic’s catalog flagship tier for reasoning, coding, and long-horizon agents; 1M context and multimodal inputs are enough; and you will pay about 2.5× Sonnet while staying clearly under Fable. Default to the regular SKU; try Fast only when latency is the bottleneck.

Skip or look elsewhere for high-volume short tasks and tight budgets (Sonnet 5 or cheaper Grok/K3); for top-tier Mythos narrative or proven answer quality that clears a further 2× premium (Fable 5); or when Opus multi-host coverage is still missing—verify Token first, or temporarily use Sonnet/Fable paths that already span more hosts.

Decide with AI Hippo’s model, Token, and compare pages; the composite board is a context-per-dollar proxy, not a capability benchmark.

Deployment checklist

Before you integrate: keep `claude-opus-5` and `claude-opus-5-fast` as distinct routes so 2× pricing never becomes the silent default; confirm the harness supports reasoning / reasoning_effort / tools / structured_outputs; budget with your real input/output ratio, not the $5 input sticker alone.

Operationally: monitor tokens per turn including reasoning; canary Fast only when latency SLAs fail; re-check the Token page for new hosts (e.g. Bedrock). Treat SQLite snapshot prices as source of truth—run `data:update` again before you ship.

Sources

• Claude Opus 5 on AI Hippo
• Claude Opus 5 Fast on AI Hippo
• Claude Sonnet 5 on AI Hippo
• Claude Fable 5 on AI Hippo
• OpenRouter models API
• Anthropic pricing
• Anthropic fast mode docs