Claude Opus 5 vs Sonnet 5: The Gap Is Exactly 2.5x
Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens. Claude Sonnet 5 costs $2 and $10. Both sides of the bill are exactly 2.5x apart.
That symmetry is unusual. In most cross-model comparisons the ratio moves with whether your traffic is input heavy or output heavy. Here it does not move: Opus 5 costs 2.5x Sonnet 5 at any mix.
Opus 5 launched on 24 July 2026 at the same price as Opus 4.8, so nothing about the switch to the new generation changed the arithmetic. Prices below were checked on 16 September 2026 against Anthropic's own pricing documentation.
AI Perks tracks $7.7M in credits across 194 companies, which is the other half of this calculation.
The Verified Anthropic Price List
Anthropic's current line runs from $1/$5 to $10/$50 per million tokens, and every tier uses the same 5x output multiple.
| Model | Input per 1M | Output per 1M | 1M in + 1M out |
|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | $60 |
| Claude Mythos 5.1 | $10 | $50 | $60 |
| Claude Opus 5 | $5 | $25 | $30 |
| Claude Sonnet 5 | $2 | $10 | $12 |
| Claude Haiku 4.5 | $1 | $5 | $6 |
Blended figures are estimates derived from the listed prices at a 1:1 token mix.
Two things fall out of that table. Output costs 5x input on every tier, so the cheapest lever on any Claude bill is shortening responses, not shortening prompts. And Opus 5 now sits at exactly half the Fable 5.1 and Mythos 5.1 price, which makes it the middle of Anthropic's own range rather than the top of it. Mythos 5.1 is limited availability.
What 2.5x Costs at Real Monthly Volumes
At 50M input and 10M output tokens a month, Opus 5 runs about $500 and Sonnet 5 about $200. The gap is $300 a month, or $3,600 a year, for one workload.
| Monthly volume | Opus 5 | Sonnet 5 | Haiku 4.5 | Opus minus Sonnet |
|---|---|---|---|---|
| 5M in / 1M out | $50 | $20 | $10 | $30 |
| 50M in / 10M out | $500 | $200 | $100 | $300 |
| 200M in / 40M out | $2,000 | $800 | $400 | $1,200 |
All figures are estimates calculated from the verified per-million prices above.
Note what happens at a 5:1 input to output ratio: because output is priced at 5x input, the two halves of the bill come out exactly equal. Teams who assume input dominates are usually wrong. The heaviest profile above carries a $14,400 annual gap between the two tiers.
Credits are what close that gap in year one, and AI Perks is where the current programs are tracked.
The Sonnet 5 Price Increase That Never Happened
Anthropic had scheduled Sonnet 5 to rise from $2/$10 to $3/$15 on 1 September 2026, then cancelled it. The $2/$10 price, first announced as introductory through 31 August 2026, is now the standard price.
Anthropic's page states it plainly: "The previously scheduled increase to $3/$15 per million input/output tokens on September 1, 2026 will not occur."
This is under-reported and it matters to anyone budgeting from an article written before September. Had the increase landed, Sonnet 5 would cost 50% more, the 50M in / 10M out profile above would run $300 instead of $200, and the Opus to Sonnet gap would have narrowed from 2.5x to roughly 1.67x. Cancelling it kept the gap wide.
If your cost model still carries a September step up for Sonnet, delete it. On the 200M / 40M profile that correction is worth $4,800 a year.
How Opus 5 and Sonnet 5 Compare Across Vendors
Opus 5 at $5/$25 undercuts OpenAI's gpt-6-astra flagship by half, and Sonnet 5 at $2/$10 is the cheapest mid tier on this table once output is counted.
| Model | Input per 1M | Output per 1M | 1M in + 1M out |
|---|---|---|---|
| OpenAI gpt-6-astra | $10.00 | $50.00 | $60.00 |
| Claude Fable 5.1 | $10.00 | $50.00 | $60.00 |
| OpenAI gpt-5.5 | $5.00 | $30.00 | $35.00 |
| Claude Opus 5 | $5.00 | $25.00 | $30.00 |
| OpenAI gpt-5.6-sol | $4.00 | $20.00 | $24.00 |
| OpenAI gpt-5.4 | $2.50 | $15.00 | $17.50 |
| OpenAI gpt-5.6-terra | $2.00 | $12.00 | $14.00 |
| Claude Sonnet 5 | $2.00 | $10.00 | $12.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $6.00 |
| Meta Muse Spark 1.1 to 1.3 | $1.25 | $4.25 | $5.50 |
| OpenAI gpt-5.4-mini | $0.75 | $4.50 | $5.25 |
| OpenAI gpt-5.4-nano | $0.20 | $1.25 | $1.45 |
| OpenAI gpt-5.6-luna | $0.20 | $1.20 | $1.40 |
| Meta Muse Spark Contributor | $0.10 | $0.20 | $0.30 |
Blended column is derived from the listed prices. Verified 16 September 2026.
Three head to head reads:
Opus 5 vs gpt-5.5. Identical $5 input, but Opus 5 charges $25 output against $30. Opus is cheaper on output heavy work.
Sonnet 5 vs gpt-5.6-terra. Identical $2 input, $10 output against $12. Sonnet 5 wins on the same mix.
Opus 5 vs gpt-5.6-sol. Sol is cheaper on both sides, $24 blended against $30, so Opus 5 costs 25% more at a 1:1 mix.
Four caveats on this table. OpenAI's gpt-6-astra, launched 3 to 4 September 2026, prices context above 200k tokens separately at $20/$75, so long context work costs double there. The legacy gpt-5, gpt-5-mini and gpt-5-nano models have an API shutdown on 10 December 2026, so do not build on them despite the low headline prices. Meta's Muse Spark Contributor tier is cheap because Meta trains on your prompts and completions, and Meta runs no startup credit program for Muse. Google Gemini prices were not verified in this round of checking, so they are not on this page rather than guessed. For xAI Grok, published token prices vary by model and sources conflict, so check xAI's own console rather than any secondhand figure.
Which Tier Fits Which Workload
Pick Sonnet 5 as the default, escalate to Opus 5 only for the specific calls where a rerun would cost more than 2.5x the token spend, and push everything routine to Haiku 4.5.
The decision is economic, not about capability rankings, which this page does not make. At a flat 2.5x, the question is whether an Opus call saves more than 1.5 extra Sonnet calls' worth of retries and review time.
Route by call, not by product. Most teams that cut their Claude bill mixed tiers inside one app rather than picking one model for everything. Classification, extraction, routing and summarisation sit on Haiku 4.5 at $1/$5. Drafting and reasoning sit on Sonnet 5. Only the hardest step hits Opus 5.
Watch the output side. With a 5x output multiple on every tier, a prompt change that cuts response length by 40% saves more than a tier downgrade on input alone.
Budget for growth, not today. The 2.5x gap compounds with volume. A prototype where Opus costs $30 more a month becomes a product where it costs $1,200 more.
Do not build on shutdown dated models. The legacy OpenAI tier looks cheap until 10 December 2026.
Cover the Bill With Credits Instead
API credits are the only thing that makes the tier choice free, and there are currently $7.7M in tracked credits across 194 companies.
A $500 a month Opus 5 workload is $6,000 a year. Credit programs routinely cover that and more, which turns the question from "which tier can we afford" into "which tier is right".
AI Perks tracks what is available across Anthropic, OpenAI, cloud providers and the infrastructure underneath them, with amounts, terms and status in one place. Model credits and compute credits are separate bills, and holding both buys a full year of runway.
Terms change often, which is why a live tracker beats a blog post you read once. See the current list at getaiperks.com.
Frequently Asked Questions
How much more expensive is Claude Opus 5 than Sonnet 5?
Exactly 2.5x on both input and output. Opus 5 is $5 per million input tokens and $25 per million output, against $2 and $10 for Sonnet 5. Because both sides share the same ratio, the 2.5x multiple holds at every possible input to output mix, which is unusual.
Did Claude Sonnet 5 get a price increase in September 2026?
No. Anthropic had scheduled a rise to $3/$15 per million tokens on 1 September 2026 and cancelled it. The $2/$10 price, originally announced as introductory through 31 August 2026, is now standard. Any cost model carrying that September step up is overstating Sonnet 5 spend by 50%.
What does Claude Opus 5 cost per month?
At 50M input and 10M output tokens a month, roughly $500, estimated from the $5/$25 per million price. Light use at 5M in and 1M out is about $50, and heavy use at 200M in and 40M out is about $2,000. Credits covering those bills are tracked at getaiperks.com.
Is Claude Opus 5 cheaper than OpenAI's flagship?
Yes, on list price. Opus 5 is $5/$25 per million tokens and OpenAI's gpt-6-astra is $10/$50, so Opus 5 is half. Astra also prices context above 200k tokens at $20/$75. Anthropic's Fable 5.1 sits at $10/$50, level with astra. Both were verified on 16 September 2026.
Which Claude model should I use for a coding agent?
Route by call rather than picking one model. Output tokens cost 5x input on every Claude tier, so long generations are where the bill lands. Run routine steps on Haiku 4.5 at $1/$5, drafting on Sonnet 5, and reserve Opus 5 for the steps where a retry costs more than the token gap.
How do I get free Claude API credits?
Multiple programs offer Anthropic credits, and they stack with cloud and infrastructure credits that cover a different part of the bill. AI Perks tracks $7.7M across 194 companies with current amounts and terms, so you can see what applies before spending a cent on tokens.
Pick the tier your workload needs. Let someone else pay the token bill.