Sonnet 5 vs Haiku 4.5 vs Opus 5.5 (Sept 2026): Which Claude Model to Use

Update (September 28, 2026): Anthropic released Claude Sonnet 5.5, replacing Sonnet 5 as the current Sonnet at the same $2/$10 API price. The recommendations below were written for Sonnet 5; for how Sonnet 5.5 compares with Opus 5.5, see Claude Sonnet 5.5 for professionals. Anthropic has not yet published which plans default to Sonnet 5.5.

TL;DR. On a paid plan, start with Opus 5.5 — the new Opus, released September 22, 2026 at a lower price than Opus 5, and Anthropic's own recommended starting point "for most workloads." Sonnet 5 is the claude.ai default, the model Free users get, and the faster, cheaper pick for quick tasks and high-volume work. Haiku 4.5 wins on speed and scale. Fable 5.1 sits above them all when maximum capability justifies its separate billing.

Updated September 22, 2026: Anthropic released Claude Opus 5.5 (claude-opus-5-5), which replaces Opus 5 as the current Opus on paid claude.ai plans. It lists at $4/$20 per million tokens (Opus 5 was $5/$25), and Anthropic says it "performs at the level of Claude Fable 5.1 on most work." This guide now reflects it; the launch details are in our Claude Opus 5.5 write-up.

Use	Pick	Why
Most professional work on a paid plan — drafting, analysis, hard code, long agentic work	Opus 5.5	Anthropic: "start with Claude Opus 5.5 for most workloads"; built for long-running agentic coding and knowledge work; $4/$20 (cache reads $0.20), 1M context
Quick tasks, the Free plan, and cost-sensitive API volume	Sonnet 5	Anthropic's "best combination of speed and intelligence"; $2/$10 per million tokens, 1M context, fast
The hardest, longest-running agent work when cost is secondary	Fable 5.1	Anthropic's most capable generally available model; $10/$50 (cache reads $0.25), 1M context, thinking always on, slower
Volume, speed, lowest cost	Haiku 4.5	Fastest model with near-frontier intelligence; $1/$5, 200K context

Anthropic's current generally-available lineup as of September 29, 2026: Sonnet 5.5 (released September 28, replacing Sonnet 5 as the current Sonnet; Anthropic has not yet published which Sonnet version each plan runs — see our Claude Sonnet 5.5 write-up), Opus 5.5 (the newest Opus, released September 22 — see our Claude Opus 5.5 write-up), Haiku 4.5, and Fable 5.1, a Mythos-class model that sits above the Opus tier (see our Claude Fable 5.1 write-up for what changed on September 1 and how it is billed). All four share a common feature surface (text + image input, vision, tool use, Skills, Projects) but differ meaningfully on capability, speed, cost, and context window. Picking the right one for the right task is still the highest-leverage tooling decision most professionals make in 2026.

This guide does the translation. Specs cited come from Anthropic's model documentation as of September 22, 2026.

The four current models at a glance

	Fable 5.1	Opus 5.5	Sonnet 5	Haiku 4.5
Position	Top tier; most capable generally available model, for when cost is secondary	Current Opus; long-running agentic coding and knowledge work	The default; best combination of speed and intelligence	Fastest; near-frontier intelligence
API ID	claude-fable-5-1	claude-opus-5-5	claude-sonnet-5	claude-haiku-4-5-20251001
Context window	1M tokens	1M tokens	1M tokens	200K tokens
Max output	128K tokens	128K tokens	128K tokens	64K tokens
Pricing (API)	$10/M input, $50/M output (cache reads $0.25/M)	$4/M input, $20/M output (cache reads $0.20/M)	$2/M input, $10/M output	$1/M input, $5/M output
Latency	Slower	Moderate	Fast	Fastest
Thinking	Adaptive (always on), effort-controlled	Adaptive (always on), effort-controlled	Adaptive, effort-controlled (no manual extended-thinking mode)	Extended thinking (manual)
Default effort (API)	high	medium	high	Not supported
Reliable knowledge cutoff	June 2026	June 2026	Jan 2026	Feb 2025

Opus vs Sonnet vs Haiku: the tiers explained

Anthropic ships Claude in three named tiers, and the names describe a capability-cost-speed trade-off rather than a specific release. Opus is the top workhorse tier: the most capable and most expensive of the three, built for hard reasoning, complex code, and long agentic work where an early mistake compounds. Sonnet is the balanced middle tier: strong on most knowledge work at a lower price and faster speed, which is why it's the default on claude.ai. Haiku is the small tier: fastest and cheapest, for high-volume, latency-sensitive, or background tasks where throughput matters more than depth.

The tier names persist across generations; the version number says which release you're on. As of September 29, 2026 the current versions are Opus 5.5, Sonnet 5.5, and Haiku 4.5 (Anthropic says Haiku 5.5 is coming "in the coming weeks", with no date). Above Opus sits a fourth tier, Fable 5.1 — a Mythos-class model Anthropic calls its most capable generally available model, for when even Opus isn't enough and cost is secondary. Price tracks the tiers: Haiku $1/$5, Sonnet $2/$10, Opus $4/$20, Fable $10/$50.

The single most important question: are you on claude.ai or the API?

If you're on claude.ai (the consumer/team chat product), Sonnet 5 has been the picker's default for Free and Pro users since July 1, 2026, but on a paid plan it's worth switching to Opus 5.5 for most professional work — it's Anthropic's own recommended starting point and costs you nothing extra per message. Opus 5.5 is available on every paid plan (Pro, Max, Team, and Enterprise) and became the current Opus on September 22, succeeding Opus 5; Free stays on Sonnet 5 and Haiku. Pro ($17/mo annual or $20/mo monthly) and above get the full picker. Switch freely as the task changes — there's no per-token cost to you, just usage limits per plan (Anthropic raised the five-hour limits on paid plans alongside the Opus 5.5 launch; see Claude pricing).

If you're using the API (or Claude Code on an API key), each request costs by token. Anthropic's models page says that if you're unsure, start with Opus 5.5 for most workloads. At $4/$20 it costs twice Sonnet 5's rate (Opus 5 was 2.5x), so for high-volume, latency-sensitive, or simple API work, Sonnet 5 ($2/$10, the cheapest 5-generation option) is still worth testing against it on your own tasks.

If you just want one default: Opus 5.5 on a paid plan, Sonnet 5 on Free. The rest of this guide is about the exceptions.

Pick Opus 5.5 for: most professional work on a paid plan — especially hard code, long agentic chains, and high-stakes work

Opus 5.5 is Anthropic's newest Opus-tier model, released September 22, 2026. Anthropic describes it as built "for long-running agentic coding and knowledge work," says it "performs at the level of Claude Fable 5.1 on most work," and reports that it generates output more than 30% faster than Opus 5. Where it clearly earns the step up from Sonnet 5:

• Complex software engineering — multi-file features, larger refactors, end-to-end feature work in Claude Code
• The longest multi-step agentic workflows — compliance automation, data-extraction pipelines, anything that chains dozens of steps where a single early error compounds
• Second opinions on high-stakes deliverables — when the document goes to a regulator, court, or board and you want the strongest model your plan includes to draft or check it

Two practical notes for API users. First, Fast mode on Opus 5.5 (research preview, Claude API only) costs $8/M input and $40/M output for significantly faster output. Second, the effort parameter defaults to medium on Opus 5.5 (it defaulted to high on Opus 5); raise it for the hardest work, or set it explicitly if you're migrating code that assumed the old default.

Profession-specific Opus 5.5 wins

• Data scientists and technical operators running hard multi-step pipelines in Claude Code
• Healthcare compliance officers running QSR gap audits across full DHF documents where miss-cost is extreme
• AI compliance officers producing pre-legal regulatory screens across EU AI Act tiers + Annex III + US state overlays
• ESG sustainability analysts mapping KPIs to multiple frameworks from a 200-page sustainability report

Pick Sonnet 5 for: speed, the Free plan, and high-volume work

Sonnet 5 is the picker's default and the only 5-generation model on Free, and it's faster and half Opus 5.5's API price. It also collapsed most of the old "when to pay for Opus" calculus. On real-world knowledge-work benchmarks it edged out Opus 4.8 (GDPval-AA v2, a real-world knowledge-work benchmark: 1,618 vs 1,615), and it leads on computer-use tasks (OSWorld-Verified: 81.2%). In practice that means:

• Client / borrower / customer communication — the daily back-and-forth, at speed
• Long-form structured documents — memos, disclosure drafts, PRDs, case analyses. This used to be Opus territory; Sonnet 5 handles most of it well
• Long-document analysis — full case files, medical histories, board packages, policy documents. 1M context, same as Opus
• Multi-step workflows — Sonnet 5's agentic improvements were the headline of its release; most chained workflows don't need Opus
• Iteration and polishing — fast enough to think alongside you

Profession-specific Sonnet 5 wins

• Loan officers running the four-audience pipeline update workflow daily across 8+ active loans
• Real estate agents generating listing descriptions, client emails, CMAs, market updates
• Attorneys and paralegals drafting memos and summaries from full source documents — verify the hardest analyses on Opus if the stakes demand it
• Copywriters and community managers doing iteration-heavy, voice-sensitive work
• Most healthcare clinicians drafting SOAP notes, treatment plans, patient education
• Management consultants synthesizing meeting notes into strategy memos
• AI product managers structuring specs and rollout plans

Pick Haiku 4.5 for: speed and scale

Haiku 4.5 is the fastest current Claude model and the cheapest. It's also more capable than people expect — "near-frontier intelligence" per Anthropic's framing. The places it earns its place:

• High-volume classification or extraction tasks — running over thousands of inputs per day
• Real-time chat surfaces — when the response needs to feel instantaneous (in-product chatbots, customer support assistants)
• Background AI features inside production applications — the AI that runs invisibly behind a feature, where latency directly affects user experience
• Cost-sensitive workflows where Sonnet's depth isn't worth the price — at $1/$5, Haiku is the cheapest current model, half of Sonnet 5's $2/$10

Haiku's constraints to know:

• 200K context window, not 1M. Long documents need chunking
• Manual extended thinking only — no effort-controlled adaptive thinking
• Reliable knowledge cutoff is Feb 2025 — older than Sonnet 5's (Jan 2026). For questions about events in 2025 or 2026, Haiku may have stale knowledge

Profession-specific Haiku 4.5 wins

• Customer support functions building in-product AI chat where latency matters
• Recruiters / HR running high-volume resume classification or initial screening (with appropriate human-in-the-loop and EEOC-aware guardrails — see our recruiter audit guide)
• Sales teams generating high-volume personalized outreach where the template is the value-add and depth matters less
• Internal tooling — Slack bots, internal knowledge search, the AI behind ops dashboards
• Real-time content moderation in community management workflows

Is there a Haiku 5?

No. As of September 29, 2026, Anthropic's models page lists Haiku 4.5 as the current Haiku model — the current lineup is Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 4.5. There is no "Haiku 5"; Anthropic has announced that Haiku 5.5 will arrive "in the coming weeks" but has not given a date.

What to use instead depends on why you wanted it:

• Cheapest and fastest: stay on Haiku 4.5. The models page lists it at $1/M input, $5/M output — still the lowest-cost current Claude model.
• Claude 5-generation capability at the lowest price: step up to Sonnet 5 at $2/$10. It costs twice Haiku's rate, but brings 1M context and adaptive thinking.

Head-to-head: the matchups people actually search

Sonnet 5 vs Haiku 4.5

Verdict: Sonnet 5 for quality-sensitive work, Haiku 4.5 for speed- and volume-sensitive work. If a human reads the output and judges it, use Sonnet 5; if the output feeds a pipeline, a chat widget, or a thousands-per-day loop, use Haiku 4.5.

The practical differences: Sonnet 5 has a 1M-token context window against Haiku's 200K, a newer knowledge cutoff (Jan 2026 vs Feb 2025), and effort-controlled adaptive thinking. Haiku answers faster and costs half of Sonnet 5's rate ($1/$5 vs $2/$10). For most professionals working through claude.ai, this one is easy — Sonnet 5 is the default and the right call; Haiku is a builder's model.

Opus 5.5 vs Sonnet 5

Verdict: Opus 5.5 by default on a paid plan; Sonnet 5 when speed, volume, or API cost matters more. That follows Anthropic's own guidance to start with Opus 5.5 for most workloads. Sonnet 5 is half Opus 5.5's API price ($2/$10 vs $4/$20), faster, and still handles everyday drafting and analysis well.

Opus 5.5 pulls ahead on complex software engineering, long agentic chains, and deep multi-step reasoning — Anthropic positions it at Fable 5.1's level on most work. The gap in price is narrower than it was with Opus 5 ($5/$25), which makes Opus 5.5 easier to justify for API workloads where quality matters. On claude.ai the calculus is simpler still: there's no per-token cost, so switch to Opus 5.5 whenever you hit hard code or a long agentic session and switch back after.

Opus 5.5 vs Opus 5

Verdict: Opus 5.5 — it's cheaper and faster. Opus 5.5 lists at $4/$20 against Opus 5's $5/$25 (20% lower), cache reads drop from $0.50 to $0.20 per million tokens, and Anthropic says typical workloads cost about 40% less because the model is also less verbose. Output is more than 30% faster. Both have a 1M-token context window and 128K max output.

On claude.ai, Opus 5.5 is now the current Opus on paid plans. On the API, Opus 5 stays available as a legacy model, so pinned integrations won't break — but moving to claude-opus-5-5 is not a pure model-string swap the way Opus 4.8 → Opus 5 was. Adaptive thinking is always on and can't be disabled, forced tool use isn't supported, and the default effort is medium. Read Anthropic's Opus 5.5 migration guide first; our Opus 5.5 write-up summarises the breaking changes.

Fable 5.1 vs Opus 5.5

Verdict: Opus 5.5 by default; Fable 5.1 for the most demanding reasoning and long-horizon agentic work. That is Anthropic's own guidance: start with Opus 5.5 for most workloads, and move to Fable 5.1 for demanding reasoning and long-horizon agentic work, or when Opus 5.5 at a higher effort setting still falls short.

The cost gap is 2.5x per token ($10/$50 vs $4/$20), and cache reads are now slightly cheaper on Opus 5.5 ($0.20/M) than on Fable 5.1 ($0.25/M), so Opus 5.5 wins on price in every usage pattern. On claude.ai the difference is plan access: Opus 5.5 is included on every paid plan, while Fable 5.1 draws on usage credits on Pro and counts against a 50% weekly share on Max.

Older matchups (previous-generation models)

These pairings still get asked about; the models remain available via the API even though newer options exist.

Opus 5 vs Sonnet 5

The question from July to September 2026. Opus 5 ($5/$25) led on hard code and long agentic chains while Sonnet 5 covered most knowledge work at less than half the price. Opus 5.5 replaced Opus 5 on September 22 at a lower price, so the hard-work answer is now Opus 5.5.

Opus 4.8 vs Sonnet 5

The pre-Opus-5 question. Sonnet 5 matched Opus 4.8 on knowledge work while Opus 4.8 led on complex software engineering (SWE-bench Pro: 69.2% vs 63.2%). The answer was "Sonnet 5 unless it's hard code" — and the hard-code answer has since moved to Opus 5, then Opus 5.5.

Opus 4.8 vs Sonnet 4.6

Both are previous-generation now. Opus 4.8 was the stronger model across the board — Sonnet 4.6's case was price. Today the same money buys strictly better options: Sonnet 5 costs less than Sonnet 4.6 did and beats it decisively, and Opus 5.5 costs less than Opus 4.8 did.

Haiku 4.5 vs Sonnet 4.6

Haiku 4.5 remains current; Sonnet 4.6 is legacy. If you're choosing between these two today, the real choice is Haiku 4.5 vs Sonnet 5 — Sonnet 5 is better and cheaper than Sonnet 4.6 ever was. Haiku still wins wherever latency and volume dominate.

Where does Fable 5.1 fit?

Fable 5.1 (claude-fable-5-1, released September 1, 2026) is a Mythos-class model that sits above the Opus tier entirely. It is Anthropic's most capable generally available model; Claude Mythos 5.1 is the same model with fewer safeguards, offered only to vetted organizations. Per-token pricing is unchanged from Fable 5 at $10/$50, but cache reads dropped from $1 to $0.25 per million tokens — Anthropic says that alone cuts typical workload cost by around 25%, and up to about 45% on heavily agentic work. Fable 5 remains available as a legacy model at the same per-token price but with the old $1 cache-read rate, so there is no reason to pick it for new work. Fable 5.1 and Opus 5.5 share a June 2026 reliable knowledge cutoff.

The catch is still billing. Fable models are not included in Pro's plan limits — Pro users run them on pay-as-you-go usage credits at API rates. Max includes Fable 5.1 up to 50% of your weekly usage (Team and Enterprise Premium seats work the same way), and Free gets no access. API use requires accepting a 30-day data-retention term for safety monitoring, and Claude Code needs v2.1.255 or later to select it.

Practical rule: treat Opus 5.5 as the workhorse ceiling — Anthropic says it performs at Fable 5.1's level on most work — and reserve Fable 5.1 for the sessions where the deepest reasoning or the longest agentic runs are worth paying for separately. For the full story, see our Claude Fable 5.1 write-up; the earlier Claude Fable 5 explainer covers the June launch and the export-control suspension.

The decision tree

If you only remember one decision rule, use this:

1. Are you on a paid plan doing professional work — drafting, analysis, documents, hard code, agentic chains? → Opus 5.5. Anthropic's recommended starting point
2. Are you on Free, or is it a quick task where speed matters more than depth (or high-volume API work where cost does)? → Sonnet 5
3. Is latency the user-facing experience, or is this running at thousands-per-day volume? → Haiku 4.5
4. Is this the rare session where Opus 5.5 still falls short and your plan's Fable usage (or credits) covers it? → Fable 5.1

When to switch mid-conversation

On claude.ai you can switch models mid-conversation. The pattern that works for serious work:

• Stay in Sonnet 5 for thinking, drafting, iterating — and for most final deliverables too
• Switch to Opus 5.5 when you hit genuinely hard code, a long multi-step agentic task, or a deliverable where you want the strongest model your plan includes
• Switch to Fable 5.1 for the occasional session where maximum capability is worth it and your plan allows

The old discipline — "draft in Sonnet, finish in Opus" — still works, but it's now reserved for the work that genuinely needs it rather than applied to everything.

Effort levels: how Opus 5.5 and Sonnet 5 tune depth

Opus 5.5, Sonnet 5, and Fable 5.1 all take an effort parameter that controls how much the model deliberates before answering. Lowering it trades depth for speed and cost; raising it suits long-running agentic and coding sessions. The API defaults differ: medium on Opus 5.5, high on Sonnet 5 and Fable 5.1 (Anthropic's effort docs list the available levels).

The question that used to come up constantly — "Opus on low effort vs Sonnet on high?" — has a cleaner answer now:

• For knowledge work, Opus 5.5 at its default medium effort is the recommended starting point; Sonnet 5 at default effort is the faster, half-price alternative for quick tasks.
• For hard code and long agentic chains, Opus 5.5 wins on base capability. Its default medium effort is a sensible starting point; raise it for the hardest work before reaching for Fable 5.1, which is what Anthropic's own guidance suggests.
• Rule of thumb: pick the model by the task's type (knowledge work → Sonnet 5; hard technical or high-stakes work → Opus 5.5), then use effort to tune speed and cost within that model.

(Haiku 4.5 is the exception: it exposes manual Extended Thinking, which you invoke explicitly, rather than effort-controlled adaptive thinking.)

What about the legacy models?

Anthropic still lists Fable 5, Opus 5, Opus 4.8, Opus 4.7, Opus 4.6, Opus 4.5, Sonnet 4.6, and Sonnet 4.5 as legacy models that remain available. Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) are deprecated and retired as of June 15, 2026. Opus 4.1 was retired on August 5, 2026.

Opus 5 is now the previous-generation Opus. It remains available on the API, but Opus 5.5 is cheaper ($4/$20 vs $5/$25) and faster — check the migration guide for the breaking changes before you switch. Opus 4.8 and Sonnet 4.6 are legacy too: if your integration pins claude-sonnet-4-6, update it to claude-sonnet-5 — a better model at a lower rate — and if it pins claude-opus-4-8, Opus 5.5 is both better and cheaper.

How API model versioning works

One subtle change starting with Claude 4.6: model IDs are pinned snapshots, not evergreen pointers. claude-opus-5 won't auto-upgrade to Opus 5.5 — it stays on its snapshot, exactly as claude-opus-4-8 stayed put when Opus 5 launched. This is a real change from earlier versioning patterns where some aliases moved.

For production integrations, this is the right behavior: you don't want your model silently changing under you. For staying current, it means actively migrating model strings when new versions ship — as everyone pinned to claude-opus-5, claude-opus-4-8, or claude-sonnet-4-6 should be planning now.

Pricing context

Per-token pricing as of September 22, 2026 from Anthropic's documentation:

• Opus 5.5: $4/M input, $20/M output; cache reads $0.20/M (5% of the input price); Fast mode $8/$40; Batch $2/$10. Anthropic says typical workloads cost about 40% less than on Opus 5
• Opus 5 (legacy): $5/M input, $25/M output; cache reads $0.50/M
• Sonnet 5: $2/M input, $10/M output — the launch rate is now the standard price (Anthropic cancelled the increase to $3/$15 it had scheduled for Sept 1, 2026). Updated tokenizer produces 1.0–1.35x more tokens for the same content vs Sonnet 4.6
• Haiku 4.5: $1/M input, $5/M output
• Fable 5.1: $10/M input, $50/M output; cache reads $0.25/M. Fable 5 stays available as a legacy model at the same per-token price but with $1/M cache reads, so there is no reason to pick it

For most professionals working through claude.ai rather than the API, per-token pricing is academic — usage limits are per-plan, not per-token. But if you're using Claude Code on an API key, or you're integrating Claude into your own product, the per-token pricing is where the per-model cost calculus lives.

Anthropic's pricing page has the current consumer plan details. Verify before committing to a plan.

Bottom line

On a paid claude.ai plan, switch the picker to Opus 5.5 — it's Anthropic's recommended starting point for most workloads and costs you nothing extra per message. Use Sonnet 5 on Free or when speed matters more than depth; Haiku 4.5 when speed or volume is the point; and Fable 5.1 for the rare session where Opus 5.5 falls short and maximum capability justifies its separate billing.

The era of "just use Opus for everything" is over — but with Opus 5.5 at $4/$20, the premium for reaching for it when it matters is smaller than it has ever been. Pick the model for the task, not the task for the model.

Model is only half the decision. The other half is surface — chat, Cowork, Claude Code, or the Chrome extension — and picking the wrong one costs more than picking the wrong model: see which Claude surface for which task, or answer four questions in the surface chooser.

For the deeper dives, see our Claude Opus 5.5 write-up, the earlier Claude Opus 5 write-up, the Claude Fable 5.1 write-up, the earlier Claude Fable 5 explainer, and the Claude Sonnet 5 launch write-up. For Claude vs ChatGPT, see our profession-specific comparison hub.

---

This article cites model specifications as published in Anthropic's model documentation, pricing as published in Anthropic's pricing documentation and at claude.com/pricing, and Opus 5.5 launch claims from Anthropic's Opus 5.5 announcement, as of September 22, 2026. Anthropic updates model availability, capabilities, and pricing frequently. Verify current state before procurement or integration decisions. Plan defaults and launch dates come from Anthropic's announcements and help center; benchmark and cost-reduction figures are Anthropic's published results.