On 24 July 2026, Anthropic released Claude Opus 5 (model id claude-opus-5). Official positioning: capability close to Claude Fable 5, at half Fable’s price — the same list price as Opus 4.8, $5 input / $25 output per million tokens.
For freeaiapi readers, a new model is good news — but equally important:
Is there a simple free API you can actually use today?
Bottom line: the most reliable free API path right now is still Vercel AI Gateway’s $5 monthly credits. Rate limits are strict, so it is not production-ready, but it is fine for day-to-day testing. Anthropic’s official API, OpenRouter, and Claude Pro/Max are not free. If the model looks good after testing and you want stable production use, try our sister site MuiRouter — faithful routing, low fees. Their technical write-up also covers capability and pricing clearly; recommended reading.
Last verified: 25 July 2026. Free credits and free-tier eligibility can change; re-check the official links before you rely on them.
Launch notes: same price, new generation (no price hike for the rename)
From the Anthropic announcement and models overview:
- Model id:
claude-opus-5(API / Claude Code / Claude Platform) - New default on Claude Max; also the strongest model currently available on Claude Pro
- Pricing: $5 input / $25 output per 1M tokens (same as Opus 4.8)
- Context: 1M tokens; max output: 128k tokens
- Optional Fast mode: about 2.5× default speed at 2× base price
- Focus: long-horizon agents, software engineering, knowledge work; stronger self-check, verification, and iteration
- Safety: alignment called among the best recently; vulnerability *finding* approaches Mythos, but full exploit *authoring* is intentionally weaker
Official messaging no longer obsesses over “price per token.” It wants you to believe the new model finishes the same class of tasks at lower total cost. Frontier-Bench, CursorBench, OSWorld, AutomationBench, and related curves all push that story.
List prices (API)
claude-fable-5: $10 input / $50 output per MTokclaude-opus-5: $5 / $25claude-opus-4-8: $5 / $25claude-sonnet-5: promo $2 / $10 through 2026-08-31, then $3 / $15
Still on Opus 4.8? Prefer a same-price upgrade and compare pass rate plus token use on your own harness — switching is often fine. Reach for Fable only when you need extreme long-horizon work and can afford the budget and guardrails; use Sonnet 5 for high-volume, more templated steps.
Early community take: value is the theme; harness decides the feel
One day after launch, public discussion is mixed — expected when agent harnesses differ by a full tier.
OpenResty author Chun (Zhang Yichun) rates Opus 5 highly; in his framework it can even beat Fable. Indie developer Baye is cooler: less depth, more human management required.
Recurring themes elsewhere:
- “Half-price Fable alternative”: API list Fable ≈ $10 / $50 vs Opus 5 $5 / $25; CursorBench claims max effort within ~0.5% of Fable. For team budgets, that is a re-price, not a rename.
- “Self-checks and self-fixes”: opening a browser at desktop/phone widths, finding root causes community patches missed — fewer nudge rounds, less rework.
- “Pushes back without empty stubbornness”: rejects a bad design, names the single disagreement, proposes a compromise — judgment often saves more human time than blind obedience.
- Skeptics: leaderboard #1 ≠ your contract parsing / trading path / multi-repo refactor is safer; Fast mode trades 2× list price for speed and must be costed per task.
One line: Opus 5 is the “default Opus” most worth evaluating first; whether it beats Fable for you needs your own task set.
Deeper capability notes: MuiRouter · Claude Opus 5.
Free trial: Vercel is the main path
1. Vercel AI Gateway — $5 / month for real API smoke tests
Claude Opus 5 is on AI Gateway. Model id:
anthropic/claude-opus-5
The model page states unpaid users get $5 credits every 30 days. AI Gateway bills at upstream list prices with no markup. Credits usually start after the team’s first AI Gateway request.
Why we treat this as the main free trial path:
- No Anthropic prepay and no Anthropic key required (non-BYOK path)
- Unified API / AI SDK — drop into apps or coding-agent config
- Same list price as official: Opus 5 is $5 / $25, so $5 monthly credit is only enough for short tests and integration checks, not long full-load agents
- Setup guide on this site: Vercel AI Gateway API key guide
AI SDK example:
import { streamText } from 'ai'
const result = streamText({
model: 'anthropic/claude-opus-5',
prompt: 'Implement the multi-file refactor described in this issue.',
reasoning: 'medium',
})cURL (OpenAI-compatible gateway shape; confirm the endpoint in your dashboard):
curl https://ai-gateway.vercel.sh/v1/chat/completions \
-H "Authorization: Bearer $AI_GATEWAY_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-opus-5",
"messages": [{"role":"user","content":"Summarize this PR in 5 bullets."}]
}'Limits to remember:
- Once a team buys extra AI Gateway Credits, it usually leaves free tier and loses the recurring $5 grant — keep a separate Hobby team for prototypes, not mixed with production paid accounts
- Official docs also say free tier covers a subset of models; if Opus 5 is blocked or asks for purchase, trust the dashboard and the Free Tier model list
- Output is expensive; $5 burns fast — start with medium effort and small tasks
- Like Fable, safety classifiers may block some cybersecurity-related requests; Gateway supports model fallbacks (e.g.
anthropic/claude-opus-4.8oranthropic/claude-sonnet-5)
2. No other free API for now
- Anthropic official API: billed at $5 / $25; no long-term free Opus 5 developer tier
- Claude Free web chat: Sonnet-class product experience — not the Opus 5 API
- Claude Pro / Max: product access to Opus 5 (Max even defaults to it), but subscription product quotas, not a free backend key
- OpenRouter and similar aggregators: model listed, metered paid, not a stable free tier
- One-shot signup promos / shady relays: not recommended here (same bar as our GPT-5.6 free-usage article)
If you need one key across many upstreams, with native Messages plus OpenAI-compatible chat, see sister site MuiRouter.
How to try it
1. Zero-cost API smoke test: sign up on Vercel, follow the key guide, set model anthropic/claude-opus-5, burn the $5 end-to-end 2. Capability and cost curves: read the MuiRouter post and the Anthropic announcement 3. Deep product use: Claude Pro / Max 4. Stable production: first-party API, cloud hosts, or a unified gateway such as MuiRouter
Bottom line
Claude Opus 5 is a same-price generation jump: official data and early community feedback both point to higher coding / knowledge-work ceilings and lower cost per completed task. Chun’s and other frontline engineers’ praise is worth attention — still validate on your harness.
Free API trial remains narrow: the main path is Vercel’s $5 per month. Enough to verify and demo, not an unlimited agent subsidy. Spend the credit on integration and evals first; for production, pick Opus / Sonnet / Fable from the real bill.