Thread

Respond to this Pin

SWE-bench Pro (agentic coding)%

Claude Sonnet 4.658.1
Claude Opus 4.869.2
Claude Sonnet 563.2

Anthropic launch table; Sonnet 5 figure also reported by TechCrunch. · source

Terminal-Bench 2.1 (agentic coding)%

Claude Sonnet 4.667
Claude Opus 4.882.7
Claude Sonnet 580.4

Anthropic launch table. · source

OSWorld-Verified (computer use)%

Claude Sonnet 4.678.5
Claude Opus 4.883.4
Claude Sonnet 581.2

Anthropic launch table. · source

Language model rubric2026 criteria

Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (Claude Sonnet 5, 30 Jun 2026): Artificial Analysis Intelligence Index 53 (max effort, index v4.1) as reported by The Decoder on 1 Jul 2026; the live AA page now shows 38 on v4.3.2. Coding null: Terminal-Bench 4.0 was not measured at launch (Anthropic reports Terminal-Bench 2.1, 80.4%). 1M-token context; introductory $2 in / $10 out per 1M (about $4 blended) through 31 Aug 2026, then made permanent on 10 Aug; AA ran at the earlier $3/$15 standard rate (price would be 3). 83.5 output tokens/s on Artificial Analysis (live page, 24 Oct 2026). Text and image input. 4 reviews, string-matched.

User Photos

Photos from readers, kept apart from the pin's sourced media. Rate what you ate or saw, and on a restaurant pin tag the dishes right on your photo.

Comments

References 588% CONFIDENCEOverall confidence: 88%How well the pin's source and references back up its dates.Weighted average of how firmly 5 references, the source included, support the pin's start and end times; a reference counts half as much for every 180 days older than the newestShow all pins at 75% confidence or better

The first entry is always the pin's source. Overall confidence is a weighted average of how firmly each reference supports the start and end times used above; a reference counts half as much for every 180 days older than the newest.

[1]90%
anthropic.com/news/claude-sonnet-5
anthropic.com· Posted Sep 28, 2026· Starts Jun 30, 2026 ✓· 24% of score
The post is dated "Jun 30, 2026" and says "From today, Claude^[2] Sonnet 5 is available across all plans"; the docs model page lists "Released June 30, 2026" and TechCrunch's^[3] 2026-06-30 report says it is the default "Starting Tuesday".
[2]90%
Claude Sonnet 5 - Claude Platform Docs
platform.claude.com· Added Sep 28, 2026· 24% of score
Anthropic's^[1][5] model page: claude-sonnet-5, a drop-in upgrade for Sonnet 4.6 with adaptive thinking on by default, 1M-token context, 128K max output (300K Batch beta), $2/$10 per million tokens, text and images in, a January 2026 knowledge cutoff, "Released June 30, 2026".
[3]88%
Anthropic launches Claude Sonnet 5 as a cheaper way to run agents
techcrunch.com· Published Jun 30, 2026· Starts Jun 30, 2026· 17% of score
TechCrunch, 2026-06-30: default for Free and Pro "Starting Tuesday"; $2/$10 per million tokens through August 31, then $3/$15, cheaper than Opus 4.8, GPT-5.5 and Gemini 3.1 Pro but dearer than Gemini 3.5 Flash; 63.2% on agentic coding against Opus 4.8's 69.2% and Sonnet 4.6's 58.1%.
[4]80%
Anthropic Launches Claude Sonnet 5 With Near-Opus Performance at a Lower Price
macrumors.com· Published Jun 30, 2026· Starts Jun 30, 2026· 17% of score
MacRumors, 2026-06-30: "Anthropic^[1][5] today introduced Claude^[2] Sonnet 5", its most agentic Sonnet, with performance similar to Opus 4.8, lower hallucination and sycophancy, and the launch pricing of $2/$10 before a planned rise to $3/$15.
[5]90%
System Card: Claude Sonnet 5
anthropic.com· Published Jun 30, 2026· 17% of score
Anthropic's^[1] system card dated "June 30, 2026", describing "the latest model in Anthropic's Sonnet family" as an upgrade to Claude^[2] Sonnet 4.6, with the behavioural audit and Firefox exploit results the launch post cites.

Suggest a correction

Something missing or wrong? Say it in your own words: a link that backs this pin up, a different start or end date and why, or a fact it lacks or gets wrong. The AI checks it against this pin's sources, searches for better ones, and adds any page that backs you up. The pin's own sources still count most. A picture that shows something else, or shows it badly, is looked at too, and moved down or replaced.

Reviews

Claude Sonnet 5 continues Anthropic's pattern of hiding price increases behind unchanged token rates · the-decoder.com · Jul 1, 2026Sonnet 5 scored 53 points at peak performance, tying with GPT-5.5 (high) for fifth place.
I tried Claude Sonnet 5 with prompts that ask it to finish the job, not just answer the question — and that's where the AI war is going · techradar.com · Jul 2, 2026On that front, Claude Sonnet 5 feels extremely capable.
Anthropic launches Claude Sonnet 5 as a cheaper way to run agents · techcrunch.com · Jun 30, 2026On a knowledge work benchmark, Sonnet 5 actually slightly outperforms Opus 4.8, which is known for winning on solving the hardest problems like making subtle judgment calls and deep research.
Anthropic's new Claude Sonnet 5 closes the gap to Opus model series · the-decoder.com · Jun 30, 2026Sonnet 5 beats its predecessor, Sonnet 4.6, across every tested category and closes in on the pricier Opus 4.8.

Sentiment over timePositive

Sentiment

Sentiment over time
Anthropic launches Claude Sonnet 5 as a cheaper way to run agents · Jun 30, 2026	0.60
Anthropic's new Claude Sonnet 5 closes the gap to Opus model series · Jun 30, 2026	0.70
Claude Sonnet 5 continues Anthropic's pattern of hiding price increases behind unchanged token rates · Jul 1, 2026	0.30
I tried Claude Sonnet 5 with prompts that ask it to finish the job, not just answer the question — and that's where the AI war is going · Jul 2, 2026	0.70

"Claude Sonnet 5 is built to be the most agentic Sonnet model yet": it plans, uses tools such as browsers and terminals and runs autonomously, with performance close to Opus 4.8 at lower prices^[1][3][4]
From launch it was the default for Free and Pro plans and available to Max, Team and Enterprise, as claude-sonnet-5 on the Claude API, with higher rate limits across Chat, Cowork, Claude Code and the Claude Platform^[1]
It launched at $2 per million input tokens and $10 per million output tokens "through August 31", with $3/$15 due to follow; on 10 August 2026 Anthropic made the introductory price permanent^[1][3][4]
It shows less misaligned behaviour than Sonnet 4.6 but more than Opus 4.8 and Claude Mythos Preview; never trained on cybersecurity, it could not build a working Firefox exploit, yet it ships with the cyber safeguards used on Opus 4.7 and 4.8^[1][5]
A drop-in upgrade for Sonnet 4.6: adaptive thinking is on by default, and manual extended thinking or non-default temperature, top_p and top_k now return an error^[2]

Notable features

Specifications: 1M-token context window, 128K max output (300K on the Batch API in beta), adaptive thinking at high default effort, text and image input with text output, and a January 2026 knowledge cutoff^[2]
Anthropic's table puts it at 63.2% on SWE-bench Pro, 80.4% on Terminal-Bench 2.1 and 81.2% on OSWorld-Verified, against Sonnet 4.6's 58.1%, 67.0% and 78.5% and Opus 4.8's 69.2%, 82.7% and 83.4%^[1][3]
On knowledge work (GDPval-AA v2) it scores 1618, just ahead of Opus 4.8's 1615 and far above Sonnet 4.6's 1395^[1][3]
A new tokenizer, like Opus 4.7's, maps the same input to roughly 1.0 to 1.35 times as many tokens^[1]
Effort levels from low to max let users trade cost for accuracy, and at high effort it matches Opus 4.8 on some tasks^[1]

Benchmarks^[1]

Benchmark	Sonnet 5	Sonnet 4.6	Opus 4.8 (for reference)
Agentic coding (SWE-bench Pro)	63.2%	58.1%	69.2%
Agentic coding (Terminal-Bench 2.1)	80.4%	67.0%	82.7%
Multidisciplinary reasoning (Humanity's Last Exam), no tools	43.2%	34.6%	49.8%
Multidisciplinary reasoning (Humanity's Last Exam), with tools	57.4%	46.8%	57.9%
Computer use (OSWorld-Verified)	81.2%	78.5%	83.4%
Knowledge work (GDPval-AA v2)	1618	1395	1615