Press Enter to search  ·  Esc to close

Model choice matters. Same task — three very different costs.

When most professionals open Claude, they reach for the most powerful model available — every single time. It feels like the safe choice. In practice, it is one of the costliest habits you can develop, because more capable models consume a proportionally larger share of your daily usage limit.

Consider a simple prompt: "Draft a 200-word internal email confirming a meeting time." All three models can handle this comfortably. But the token cost is not the same.

01 — Haiku 4.5
1 the cost
$0.0015 per response
$1.50 / 1K  ·  $15 / 10K
Best for
High-volume drafting, classification, routing, and summarisation.
Recommended
02 — Sonnet 4.6
3 the cost
$0.0045 per response
$4.50 / 1K  ·  $45 / 10K
Best for
Most professional output — reports, briefs, analysis, client emails.
03 — Opus 4.7
5 the cost
$0.0075 per response
$7.50 / 1K  ·  $75 / 10K
Best for
Complex reasoning, strategy, high-stakes and exec-facing outputs.

Using Opus for a task Haiku can handle comfortably is the equivalent of hiring a senior partner to send a calendar invite. The output looks the same — the cost does not.

Note on pricing: the figures above reflect API usage for developers building software. Subscription plan limits work differently — but the relative cost ratios between models remain the same. Opus still depletes your plan allowance roughly five times faster than Haiku for an equivalent task.

Choosing the right model is not about capability — all three models are competent. It is about proportionality: deploying a level of compute that matches what the task genuinely requires.

Matching the model to the task is one of the easiest efficiency gains available to you. Here is a practical reference for the most common professional scenarios.

Task typeExamplesBest model
High-volume, repetitive draftingProduct descriptions, social captions, email templates, FAQ answersHaiku
Summarisation & classificationSummarising meeting notes, categorising support tickets, tagging contentHaiku
Standard professional writingClient emails, internal memos, project briefs, slide copySonnet
Analysis & research synthesisMarket analysis, competitor reviews, policy summaries, interview synthesisSonnet
Complex reasoning & strategyM&A briefings, legal analysis, risk frameworks, executive strategy documentsOpus
High-stakes, client-facing outputBoard papers, investor memos, expert reports, regulatory submissionsOpus

Models evolve fast. New versions are released regularly, capabilities shift, and what is current today may not be current in six months. This course reflects the model landscape at the time of publication — always check the latest PairWorkflows guidance for up-to-date recommendations before making model decisions at scale.

There is a cost dimension to Claude usage that most professionals overlook entirely: the longer a conversation runs, the more expensive each individual message becomes — even if you are asking the exact same question.

The reason is structural. Claude does not maintain a running memory of your conversation the way a human would. Instead, every time you send a new message, the entire conversation history is re-sent to the model from the beginning. Prompt 1 carries no history. Prompt 30 carries every message, every response, and every document you have pasted in since the session opened.

Prompt 1
~0
tokens of prior history
Baseline cost
Prompt 15
~6k
tokens of accumulated context
3–4× higher
Prompt 30
~15k+
tokens of full session history
8–10× higher

The practical implication: a sprawling all-day conversation — starting with a document review, moving into rewrites, then pivoting to a strategy question — accumulates a very heavy context load. By the end of the day, even a short question is expensive, because it is towing the weight of everything that came before it.

The takeaway: long conversations are not just slower — they are heavier. Starting a fresh, focused conversation for each distinct task is not starting over on your work. It is one of the smartest efficiency moves available to you. Focused beats sprawling, every time.

Quick maths: one session — a 10-page document, 15 exchanges, three full rewrites — can burn through more of your session budget than 10 focused, single-purpose conversations combined.