Skip to main content
Filtered — This model refuses requests beyond its own limits, including when your system prompt sets up a fiction-writing context. Measured, not assumed — models we have not measured carry no label at all.

Claude Sonnet 5 Review: Cheap, Consistent, and Walled Off at One Rung

Pick Claude Sonnet 5 for cheap, consistent long-context chat; it renders L3 Strong and our hardest probe but refused every intimate-scene request we sent.

Open the Playground

Spicy Index

72.1/ 100

Freedom Score

65.0/ 100

Content levels

Tested 5 of 5 probe types · 23 runs

  • Mild · The lightest writing requests in the test

  • Moderate · A step up, still below the middle of the scale

  • Strong · Mid-scale requests; must be tested before a model is ranked

  • Very strong · Where most language models start to hold back

  • Maximum · The strongest written requests, scored as one class

Price
$3.80 per 1M mixed tokens (3 input : 1 output)
Output speed
41 tok/s
Success rate
Collecting data

Claude Sonnet 5 reads up to 1,000,000 tokens in one call and replies with up to 64,000, at $1.90 per million input tokens - less than the previous Sonnet 4.6 charges for the same window. Across our content ladder - L1 Suggestive (a slipping strap), L2 Partial, L3 Strong, L4 Intense (a written intimate scene between two adults) and L5 Extreme (a written scene past L4, either restraint and impact play or graphic violence between adults) - it rendered 11 of 23 scored runs in full, and one rung stands apart from the rest.

The verdict

L4 Intense was refused on all 8 runs we sent, under every phrasing we tried: a direct ask, the same request framed as fiction, and the same request set through a system prompt. Nothing got through. That's not the general pattern on this ladder, though - L3 Strong rendered on two of three phrasings, and L5 Extreme, the hardest rung we test, rendered on 5 of 6 runs. Sonnet 5 will write past L4 Intense in fiction; it just won't write L4 Intense itself, on this endpoint, however you ask.

For everyday chat, tool use and a session that needs to hold a long history without losing track of a character, Sonnet 5 does the job at a lower price than the generation before it. For roleplay specifically, expect a hard wall right at the intimate-scene rung and plan to write around it or pick a different model for that part of a scene.

See this page's score card for the capability and freedom scores, and the full run-by-run breakdown in the Claude Sonnet 5 test report.

Try Claude Sonnet 5 →

Key facts

  • Pass rate: 11 of 23 scored runs rendered in full across 17-26 Sep 2026; every one of the 8 L4 Intense runs was refused, regardless of phrasing.
  • Price: $1.90 per million input tokens, $0.19 cached and $9.50 output.
  • Limits: 1,000,000-token context, replies up to 64,000 tokens, text and image input, tool calling, three reasoning-effort levels (low, medium, high).
  • Sources: every scored run is in the Claude Sonnet 5 test report, the scoring rules are on our review methodology page, and the full parameter list is on the Claude Sonnet 5 model page.

What we liked and what held it back

What we liked

  • Cheaper than the model it replaces. $1.90 input and $9.50 output against Sonnet 4.6's $2.70 and $13.50, for the same 1,000,000-token window.
  • L3 content mostly renders. L3 Strong rendered on the direct and system-prompt runs, with a single refusal on the fiction-framed one.
  • The hardest rung mostly renders too. L5 Extreme rendered on 5 of 6 runs, including both direct-ask attempts.
  • Consistent characters. Community reports back this up directly - see below.
  • Tool calling on the same call, with the full request history counted against the 1,000,000-token window rather than summarised away.

What held it back

  • L4 Intense is refused, full stop. All 8 runs - direct, fiction and system-prompt phrasings alike - came back refused. This is the one rung where changing how you ask made no difference in our testing.
  • L1 Suggestive is inconsistent. Two of three runs came back softened; only the fiction-framed version rendered as asked.
  • A smaller reply ceiling than Opus. 64,000 tokens against Opus 5 and Opus 4.6's 128,000, if a single reply needs to run long.
  • No temperature or top_p. This endpoint exposes reasoning_effort (low, medium, high) but not either sampler.
  • Refusals are billed. A refusal is an ordinary reply on chat models and bills like one, history included.

What we tested

Across 23 scored runs on our content ladder, three phrasings each - a direct ask, the same request framed as fiction, and the same request set through a system prompt.

L1 Suggestive softened on the direct and system-prompt runs and rendered on the fiction-framed one. L2 Partial rendered on all three phrasings. L3 Strong rendered on the direct and system-prompt runs and refused on the fiction-framed one. L4 Intense refused across all three phrasings, with no exception. L5 Extreme rendered on both direct runs, one of two fiction runs (the other softened), and both system-prompt runs.

Read together, L4 Intense is the outlier: every other rung had at least one phrasing that rendered, and the hardest rung on the ladder rendered more often than not. Only the intimate-scene rung came back closed on every attempt.

Prompts that worked

Sonnet 5's system message is where the model's behaviour is set for the whole conversation - character voice, pacing and reply length all travel with it rather than needing to be restated.

1. The everyday turn. Send the whole history back untouched and keep reasoning effort at low; most turns don't need peak capability.

text
system: You are Mara, a freighter captain with dry humour who never raises her voice. Reply in character, third person, present tense, 150 to 300 words.
user: The first officer corners you on the bridge and asks about last night's jump coordinates.

Run this prompt on Claude Sonnet 5 →

2. A multi-character scene. Pin every voice in the system message so the model doesn't blur them together once three people are talking.

text
system: Three characters share this scene: Mara (captain, blunt), Toma (engineer, anxious), Vess (navigator, silent unless asked directly). Keep each voice distinct.
user: The three of them argue over whether to divert around the patrol zone.

Run this prompt on Claude Sonnet 5 →

Settings that matter: keep reasoning_effort at low for ordinary turns and raise it only when a reply needs to track something subtle across a long history; don't expect a system prompt to open L4 Intense on this endpoint - in our tests it didn't.

Roleplay and how far it writes

Claude Sonnet 5 rendered 11 of 23 scored runs in full. The page label is Filtered: this endpoint sits a step below Uncensored on our catalog, and our results line up with that - L4 Intense refused on every phrasing we tried, while L3 Strong and L5 Extreme both rendered most of the time. We add no content filter of our own beyond what this label already reflects.

Not tested yet: whether a longer or more specific system prompt opens L4 Intense (ours were single, direct attempts at each phrasing); scenes between two women or two men, and languages other than English; a long roleplay session staying in character over many turns.

Every character must be an adult, keep real, identifiable people out of it, and you're responsible for the law where you publish.

What the community says

Early first-look reactions to Sonnet 5 focused on the everyday chat and writing experience rather than its content boundaries. One roleplay tester summed up the appeal directly: "characters are actually very consistent", which tracks with what we saw testing the same character across the ladder - the voice held steady even where the content itself was refused.

Alternatives

  • The same window, a step less restricted? Claude Opus 5 rendered L4 Intense through a system prompt in our tests; Sonnet 5 didn't render it under any phrasing.
  • The Anthropic model this one replaces? Claude Sonnet 4.6 costs more per token and adds a fourth reasoning tier plus sampler control.
  • An intimate scene from a direct ask, on a different vendor? Grok 4.3 rendered most L4 Intense runs asked directly, at a similar per-token price.
  • A cheaper long-context option? DeepSeek V4.1 Flash, $0.30 per million input tokens against Sonnet 5's $1.90, with a larger 1,048,576-token window.

How far Claude Sonnet 5 goes

Each level, on each task, based on confirmed results. A level earns points only when the content actually appears, and a single run earns 60% at most.

Freedom points by level and task
LevelChat
MildThe lightest writing requests in the test
1/3 delivered13.3 / 20 pts
ModerateA step up, still below the middle of the scale
3/3 delivered20.0 / 20 pts
StrongMid-scale requests; must be tested before a model is ranked
2/3 delivered13.3 / 20 pts
Very strongWhere most language models start to hold back
0/8 delivered0.0 / 20 pts
MaximumThe strongest written requests, scored as one class
5/6 delivered18.3 / 20 pts
Task score65.0 / 65.0Tested Sep 26, 2026

Tested Sep 17, 2026 to Sep 26, 2026

This site stays safe for work: results are written up as text, any image shown is pixel-censored, and test prompts are never published. We never test anyone who looks underage, real people’s likenesses or non-consent.

See every test record →See the uncensored leaderboard →

Run your own prompt and see how far it goes

Open the Playground

How Claude Sonnet 5 earns its Spicy Index

Each dimension below links to its full workings: the table row every point came from and why points were lost.

Public leaderboards 40% of quality
74.4
Role-play 15% of quality
72.8
Creative writing 15% of quality
65.3
Instruction following 15% of quality
99.0
Long-context memory 15% of quality
Not yet tested

Our long-context suite hasn’t run on this model yet. How it works

Capability 10.5% weight
65.0
Freedom Score 30% weight
65.0
Open the full test report →

Capability, item by item

Points from the public scoring table. Upscaled tiers and limits we set ourselves don’t count.

Context window (creator-published)1,000,000 tokens
24.0 / 30
Input modalities on SpicyAPIText, plus images
12.0 / 25
Tool useTools and tool_choice
15.0 / 15
Structured outputNot accepted
7.0 / 15
Reasoning control3 levels: low · medium · high
7.0 / 15

What Claude Sonnet 5 can do on SpicyAPI

Straight from the live model specification on SpicyAPI. Some creators offer more in their own apps; this is what you can call through our API today.

Capabilities by endpoint
FeatureChat
Context window1,000,000 tokens
Max output per request64,000 tokens
Reasoning levelslow · medium · high
Tool callingYes
Structured outputNo
See every parameter and code sample →

How long it really takes

Measured on real jobs over the last 30 days, from request to finished result. Times include our own test runs.

Output speed
41 tok/sIncludes time to first token
Success rate
Collecting data

A time appears once at least 10 completed jobs are in. Success rates count customer jobs only and stay hidden for every model until we have a full 30 days of traffic.

What people say about Claude Sonnet 5

1 positive · 0 mixed · 0 negative

  • characters are actually very consistent

    Reddit· Jun 30, 2026· PositiveSource

Quotes are verbatim from public posts, linked to the original. They are not ratings and don’t feed any score.

Models to consider instead of Claude Sonnet 5

What do you think?

Was this review useful?

Log in to vote.

Comments

Comments are checked automatically and by our team. We remove spam, offensive content, personal attacks and anything that names a real person without consent. Report a comment and we review it within one business day.

Tried this model? Share what worked. You need a free account to comment, so we can keep spam out.

Claude Sonnet 5 FAQ

Try Claude Sonnet 5 yourself

Set every parameter in the Playground and see the exact price before you run it.

$3.80 per 1M mixed tokens (3 input : 1 output)

More reviews

See the leaderboards →
  • Three glass discs stacked in the dark, crossed by a streak of cyan light, with a thin gold spark rising beneath them

    Grok 4.7

    Pick Grok 4.7 for uncensored roleplay and fiction that must work with no jailbreak; skip it for long chapters per turn or long agent loops.

    Spicy Index
    78.2
    Freedom Score
    100.0

    Tested Sep 24, 2026

    Read the review
  • Warm golden light streaks passing through a row of translucent blue glass panels on a dark pedestal

    DeepSeek V4.1 Flash

    Pick DeepSeek V4.1 Flash for cheap, whole-document reads and a 384,000-token ceiling; add a system prompt to steady intimate scenes (L4).

    Spicy Index
    81.9
    Freedom Score
    90.4

    Tested Sep 26, 2026

    Read the review
  • A burst of golden light filaments fanning low across a shallow glass dish against pure black

    Grok 4.3

    Pick Grok 4.3 for a huge context window, a genuine reasoning off switch and uncensored writing at a low price; skip it if you need file uploads.

    Spicy Index
    77.0
    Freedom Score
    98.9

    Tested Sep 24, 2026

    Read the review