benchr
"Which Claude is best" is the wrong question. There's no single best one — there's a best one for the job in front of you and the budget you're spending, and Anthropic's lineup is built around exactly that trade. Pick wrong upward and you pay double for accuracy you didn't need. Pick wrong downward and you babysit a model that keeps missing. This guide sorts the lineup so you land on the right tier the first time.
All prices below are the published per-million-token rates from Anthropic's pricing page, and the benchmark figures come from Anthropic's model documentation. The exact numbers shift; the shape of the decision holds. If you'd rather think in dollars-per-workload than per-model, the companion piece on price per use case runs the math.
The lineup at a glance
Four tiers you can pay for, cheapest to most capable. Read this table top to bottom as a ladder: each rung buys more capability at a higher price, and the right rung is the lowest one that clears your task.
| Model | Price /1M (in/out) | Best for | Headline benchmark |
|---|---|---|---|
| Claude Haiku 4.5 | $1 / $5 | High-volume, simple, well-scoped tasks | SWE-bench Verified 73.3 |
| Claude Sonnet 5 | $2 / $10 | The cost-first everyday default | SWE-bench Verified 89.4 |
| Claude Opus 5 | $5 / $25 | Anthropic's recommended start; hard pro coding & reasoning | No published table |
| Claude Fable 5.1 | $10 / $50 | Demanding reasoning, long-horizon agentic work | — |
The empty benchmark cells are deliberate. This guide doesn't borrow numbers from a predecessor to fill a row. The nearest published reference point for Opus 5 is the Opus 4.8 it replaced at the same price, which posted SWE-bench Verified 88.6 per Anthropic — below Sonnet 5's 89.4. This page cites no Fable 5.1 figure; its launch record is in the Fable 5.1 migration guide.
Claude Haiku 4.5 — the high-volume workhorse
Haiku 4.5 is the cheap tier, and "cheap" here doesn't mean "toy." At $1 input / $5 output per million tokens (batch $0.5/$2.5, cache reads $0.10) with a 200K-token context window, it posts SWE-bench Verified 73.3 — strong for a model at this price. That combination is what makes it the right call for work measured in millions of calls, not hundreds.
Reach for Haiku when the task is high-volume and well-scoped: classification, extraction, routing, tagging, content moderation, simple chat, the inner loop of a pipeline that fires constantly. On that kind of work the marginal quality you'd buy by moving up a tier doesn't change the outcome, and the price difference compounds fast across volume.
Where Haiku stops being the right answer is ambiguity. If the task needs the model to reason through an ambiguous situation, or to write code with non-trivial structure, you'll feel the ceiling — and that's your signal to move up to Sonnet 5. Haiku 4.5 is still Anthropic's newest Haiku, but Anthropic lists its retirement as not sooner than October 15, 2026, the shortest commitment in the current lineup, so recheck its status before you build a long-lived pipeline on it. The full picture is in the Haiku 4.5 review.
Claude Sonnet 5 — the cost-first default
Sonnet 5 replaced Sonnet 4.6 and cut the price on both sides: $2 input / $10 output per million tokens (cache reads $0.20), down from $3/$15, with a 1M-token context window and 128K output. Anthropic reports SWE-bench Verified 89.4, SWE-bench Pro 71.8, GPQA Diamond 92.0 and Terminal-Bench 2.1 85.6. It's also the default model for Free and Pro plans in Claude.ai, per Anthropic's launch announcement.
The case for Sonnet 5 as your default is mostly economic. It costs two-fifths of Opus 5 on both input and output and a fifth of Fable 5.1, and its Anthropic-reported SWE-bench Verified 89.4 sits above the 88.6 Anthropic reported for Opus 4.8, the model Opus 5 replaced. So run Sonnet 5 by default and escalate the specific prompts that stump it, rather than paying a flagship rate on every call. Anthropic's own advice points the other way — start with Opus 5 — and that's the safer call when a miss costs more than the price gap. The launch record is in the Sonnet 5 launch piece.
Start on Sonnet 5. Measure where it actually fails. Then pay for Opus 5 only on the tasks that earned it — not on the whole workload.
Claude Opus 5 — Anthropic's recommended starting point
Opus 5 replaced Opus 4.8 on July 24, 2026 at the same price: $5 input / $25 output per million tokens (cache reads $0.50), with a 1M-token context window and 128K output. Anthropic's models overview names it as the model to start with for most workloads when you're unsure which to pick. What this guide can't give you is a benchmark row: Anthropic has published no benchmark table for Opus 5 that benchr records.
This is the tier to choose when an answer being wrong costs you real time or real money — hard production coding, multi-step debugging, agentic workflows where a subtle mistake propagates, dense reasoning where you'll pay for the accuracy. Be clear about what you're paying for: on published numbers, the case for Opus 5 over Sonnet 5 rests on Anthropic's positioning and your own evals, not on a benchmark gap.
Moving from Opus 4.8 costs nothing extra per token, but test before you switch. The Opus 5 review covers what to check.
High-volume simple
Haiku 4.5 $1/$5 per 1MCost-first default
Sonnet 5 $2/$10 per 1MHard pro coding
Opus 5 Pay for accuracyLong-horizon agents
Fable 5.1 Top tier you can runUnsure where to start
Opus 5 Anthropic's own adviceGated tier
Mythos 5.1 Not for saleClaude Fable 5.1 — the top tier you can use
Fable 5.1 replaced Fable 5 on September 1, 2026 at the same $10 input / $50 output per million tokens, with a 1M-token context window and 128K output. The change that moves the bill is caching: cache reads fell to $0.25 per million from Fable 5's $1.00, so a cached read now costs 2.5% of base input instead of 10%. Anthropic points to Fable 5.1 for demanding reasoning and long-horizon agentic work. At double Opus 5's base rate, it's priced for work that justifies it.
Two contract changes to check before you switch. The API returns HTTP 400 if tool_choice is set to any or to a named tool, so agents that force a tool call need rework. And Fable 5.1 isn't offered on Priority Tier. Fable 5 also returned an explicit refusal on safety-classified cyber, bio and distillation requests, with no automatic fallback to another model unless your application configured one; test that refusal path again after migrating. The full checklist is in the Fable 5.1 migration guide.
Claude Mythos 5.1 — the one you can't buy
Mythos 5.1 arrived on September 1, 2026, the same day as Fable 5.1, at a listed $10/$50 — and it's the one to set expectations on early: it's not generally available. Access is gated under Project Glasswing, so it isn't something you can sign up for or route to through the API. Treat it as out of reach for normal use.
That's why Fable 5.1 matters. If you've been waiting for the top of the lineup to open up, Fable 5.1 is as high as you can go today.
How to choose, in four questions
Skip the spec sheets and answer these in order. The first one that fits is your model.
Classification, extraction, routing, simple chat at scale → Haiku 4.5. Cheapest tier, and it clears the work.
Real coding, reasoning, drafting, agents — but not your hardest case → Sonnet 5. The cost-first default at $2/$10.
Hard production coding, multi-step debugging, dense reasoning, or no time to evaluate → Opus 5, the model Anthropic tells you to start with.
Demanding reasoning and long-horizon agentic work → Fable 5.1, as long as your agents don't force tool calls. Mythos 5.1 isn't for sale.
The mix that holds for most readers: make Sonnet 5 your default if you're watching cost, keep Haiku 4.5 wired in for the high-volume inner loops, and send hard or high-stakes work to Opus 5 — or make Opus 5 the default outright, as Anthropic advises, if the bill allows. Reserve Fable 5.1 for the long-horizon agentic runs that need it. That mix covers most work at a fraction of the cost of running a flagship on everything.
And don't over-buy. The most common Claude mistake in 2026 isn't picking a model that's too weak — it's paying flagship rates on a workload Sonnet 5 would have handled. Start one rung lower than your instinct, watch where it actually fails, and let the failures, not the fear of them, decide when you move up.
Legacy Claude models: still available, and what replaced them
Anthropic lists these as legacy models. They still answer API calls, but each has a current successor, and at least one has a retirement date close enough to plan around.
| Legacy model | Replaced by | What changes |
|---|---|---|
| Claude Fable 5 | Claude Fable 5.1 | Same $10 / $50 base price; cache reads fall from $1.00 to $0.25 per 1M; a forced tool_choice now returns HTTP 400 |
| Claude Opus 4.8 | Claude Opus 5 | Same $5 / $25 price; Opus 4.8's SWE-bench Verified 88.6 has no published Opus 5 counterpart |
| Claude Opus 4.7, 4.6, 4.5 | Claude Opus 5 | Move to Opus 5 and retest prompts before switching |
| Claude Sonnet 4.6 | Claude Sonnet 5 | $3 / $15 → $2 / $10 per 1M; SWE-bench Verified 79.6 → 89.4 |
| Claude Sonnet 4.5 | Claude Sonnet 5 | Anthropic lists retirement not sooner than September 29, 2026 |
Opus 4.8 also offered an optional fast mode at $10/$50 per million tokens for roughly 2.5x output speed. This guide has no record of whether Opus 5 carries the same option, so check Anthropic's pricing page before you plan around it. For the current lineup, Anthropic commits to not retiring Fable 5.1 before September 1, 2027, Opus 5 before July 24, 2027, Sonnet 5 before June 30, 2027, or Haiku 4.5 before October 15, 2026. The records of the models this guide used to recommend stay up: the Sonnet 4.6 review, the Opus 4.8 review and the Fable 5 launch piece.