Anthropic Launches Sonnet 5.5, Slashing Speed and Cost While Powering Its Free Tier

Anthropic announced Claude Sonnet 5.5 today, positioning it as a faster, cheaper successor to Sonnet 5 that already outperforms its predecessor on every benchmark. The company claims the new model "runs 30%+ faster, and costs up to 30% less for most work" while keeping the same price as Sonnet 5.

Key points

• Sonnet 5.5 delivers 30%+ speed gains and up to 30% lower cost per token, matching Sonnet 5 pricing.
• The model inherits a token‑limit bug from Opus 5.5, cutting off after 128,000 tokens for high‑effort requests.
• Anthropic has moved Sonnet 5.5 into its free tier, outpacing OpenAI’s Luna 5.6 in capability.

What happened

On September 28, 2026, Anthropic released Sonnet 5.5 and immediately shifted it to the free tier on claude.ai. The announcement highlighted benchmark performance, citing that Sonnet 5.5 outperforms Sonnet 5 on every metric. Testers noted that the model responds to a "max" thinking‑effort request for 128,000 tokens at a cost of $1.28 before running out of tokens and failing to generate an SVG. A separate high‑effort request (“xhigh”) produced a pelican illustration at 5.74 cents and 41 seconds.

The backstory

Sonnet 5 launched earlier this year as Anthropic’s flagship model, succeeding the Opus line. Opus 5.5, released earlier, suffered from a token‑limit bug that caused it to stop generating output after a 128,000‑token cap. Anthropic’s shift of Sonnet 5.5 to the free tier mirrors OpenAI’s strategy of offering Luna 5.6 for free users, but with a higher‑capability model.

Why speed and cost matter

Inference latency and token cost directly affect developer experience and operational budgets. A 30% reduction in latency can shorten user wait times, improve conversational flow, and enable real‑time applications such as interactive web interfaces. Lower token cost translates to smaller bills for the same workload, allowing startups to experiment more freely and scale without hitting budget ceilings. The combination of speed and cost positions Sonnet 5.5 as a compelling alternative for teams that prioritize performance without premium pricing.

Competitive implications

By moving Sonnet 5.5 to the free tier, Anthropic gains a competitive edge over OpenAI’s Luna 5.6, which remains less capable. This shift may accelerate adoption among developers who rely on free access to test and prototype. However, the lingering token‑limit bug could deter enterprise users who need consistent, high‑volume generation. The announcement also signals Anthropic’s intent to release Haiku 5.5 soon, potentially further narrowing the gap with GPT‑6 Luna.

What builders and businesses should do

• Benchmark Sonnet 5.5 against your current LLM to quantify speed and cost savings.
• Test the token‑limit behavior with your typical workload to assess reliability.
• Consider migrating low‑cost or prototyping workloads to the free tier while keeping critical pipelines on paid plans.
• Keep an eye on Haiku 5.5’s release for additional cost‑effective options.

The bottom line

Anthropic’s Sonnet 5.5 offers measurable speed and cost advantages, and its free‑tier deployment could reshape LLM adoption. Monitor Haiku 5.5’s rollout and the resolution of the token‑limit bug to determine long‑term viability.