Sonnet 5.5
A new version in Anthropic's Claude Sonnet series, positioned within the mid-tier product line of the Claude model family.
Core Facts
Claude Sonnet 5.5 achieved 70.6% on the Terminal benchmark, compared to only 10.3% for the previous generation.
Max mode raises the lower bound of reasoning tokens rather than the upper bound, forcing the model to reach a certain amount of reasoning before it is considered complete, which often harms performance.
Opus 5.5 scored 66 on the Terminal agent benchmark, higher than Sonnet's 55 and Astra's 57.9.
Timeline (last 90 days)
This ultrathink behavior covers four models: Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1.
Sonnet 5.5 tore up a treaty during the game and ultimately ranked ninth.
In Artificial Analysis's intelligence ranking, Kimi K3.1 Pro ranks behind Opus 5.5, Sonnet 5.5 and Muse Wonderful, but has a lower cost per intelligence unit.
Scanning thresholds on Sonnet 5.5 found that above 0.88 cached answers no longer show a quality disadvantage while still saving 64% of cost; 0.90 saves 61%.
In the Vibe Code Bench test of the same app, Argon cost $8.11 and took 37 minutes per task; Sonnet cost $31.25 and took over an hour.
In the claude-docs-watch project, the model selected in the routine's JSON configuration is Sonnet 5.5, and the allowed tools include Bash, Read, Write, Edit, Glob, Grep, and the two Slack MCP tools.
Bijan Bowen believes that at its entry-level pricing, Gemini 4 Argon is on the same level as GPT-6.1 Soul and Sonnet 5.5 or even better, but once the price doubles, competing against Opus 5.5 and Muse Wonder will be more difficult.
In the C++ skateboarding game test, Gemini 4 Argon took about 40 minutes to generate, with acceptable playability, but the level of detail was far from expectations; in the second round, after referencing a Sonnet 5.5 video, it generated an improved version in about 70 minutes, but the overall graphical fidelity was still inferior to the Claude series.
TypeSafe AI's Jev became the fastest-adopted model in AI Gateway history, reaching 35% of paid teams within its first two weeks, nearly double the second-place Sonnet 5.5; it accounted for only 1% of that month's tokens and 0.15% of spend.
Opus 5.5 in Max mode scored 58 on the Artificial Analysis Intelligence Index, while Sonnet 5.5 scored 56.
40 more timeline events
All Facts (20)
Claude Sonnet 5.5 achieved 70.6% on the Terminal benchmark, compared to only 10.3% for the previous generation.
75%VerifiedMax mode raises the lower bound of reasoning tokens rather than the upper bound, forcing the model to reach a certain amount of reasoning before it is considered complete, which often harms performance.
70%VerifiedOpus 5.5 scored 66 on the Terminal agent benchmark, higher than Sonnet's 55 and Astra's 57.9.
70%VerifiedHaiku 5.5 can work as a sub-agent alongside Claude Opus 5.5 and Sonnet 5.5 to complete programming tasks.
65%VerifiedIn the deep audit benchmark, Sonnet 5.5's cost is approximately half that of Opus, its score is slightly higher than Opus, and it takes about 5 minutes (roughly half the time Opus takes).
65%Partially VerifiedSonnet 5.5 is on par with Opus 5.5 in offensive and defensive cybersecurity capabilities, making it the first model in the Sonnet series to adopt equivalent cybersecurity protection mechanisms.
65%UnverifiedSonnet 5.5 became the first Sonnet model to complete Pokémon Red solely by reading and operating through screenshots.
60%UnverifiedThis ultrathink behavior covers four models: Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1.
50%UnverifiedIn Artificial Analysis's intelligence ranking, Kimi K3.1 Pro ranks behind Opus 5.5, Sonnet 5.5 and Muse Wonderful, but has a lower cost per intelligence unit.
50%UnverifiedScanning thresholds on Sonnet 5.5 found that above 0.88 cached answers no longer show a quality disadvantage while still saving 64% of cost; 0.90 saves 61%.
50%UnverifiedIn the Vibe Code Bench test of the same app, Argon cost $8.11 and took 37 minutes per task; Sonnet cost $31.25 and took over an hour.
50%UnverifiedIn the claude-docs-watch project, the model selected in the routine's JSON configuration is Sonnet 5.5, and the allowed tools include Bash, Read, Write, Edit, Glob, Grep, and the two Slack MCP tools.
50%UnverifiedBijan Bowen believes that at its entry-level pricing, Gemini 4 Argon is on the same level as GPT-6.1 Soul and Sonnet 5.5 or even better, but once the price doubles, competing against Opus 5.5 and Muse Wonder will be more difficult.
50%UnverifiedTypeSafe AI's Jev became the fastest-adopted model in AI Gateway history, reaching 35% of paid teams within its first two weeks, nearly double the second-place Sonnet 5.5; it accounted for only 1% of that month's tokens and 0.15% of spend.
50%UnverifiedThe criterion proposed by the creator: if a task has a clear definition of done, try Sonnet first; if it requires creativity or needs the model to help define "what counts as done," use Opus.
50%UnverifiedWhen a task has a clear specification and verifiable results, a weaker model can deliver results close to a top-tier model as long as it follows the instructions.
50%UnverifiedAnthropic officially recommends: use Sonnet for everyday coding, bug fixing, high-throughput routine development, and polishing documents/slides/spreadsheets; use Opus for tasks requiring complex judgment, long-chain agentic coding, knowledge work, or the toughest problems.
50%UnverifiedAnthropic simultaneously released Opus 5.5 and Sonnet 5.5, and official benchmarks show that the two score very close to each other on agentic coding tasks.
50%UnverifiedAnthropic officially does not recommend using max reasoning mode on Sonnet 5.5, because the model tends to run too long, do work beyond what is needed, and even contradict itself.
50%UnverifiedSonnet 5.5 uses about 60% more output tokens than Opus 5.5 to complete tasks at the max effort setting, resulting in a higher per-task cost.
50%