Claude Sonnet 5.5 Is Now Free on Felo Search: Near-Opus Answers at Sonnet Price
Claude Sonnet 5.5 is live in Felo Search: 70.6% on Terminal-Bench 4.0, a 1M-token context, and $2/$10 pricing. Here is what changes for your research.
Anthropic released Claude Sonnet 5.5 on September 28, 2026, and the headline number is not the price. It is 70.6% on Terminal-Bench 4.0 — four points ahead of Claude Opus 5.5, a model that costs twice as much.
Sonnet 5.5 is live in Felo Search now. It is the Sonnet-tier model, priced at $2 per million input tokens and $10 per million output, and it closes most of the gap to the flagship while beating it on agentic coding. For a search engine that runs a multi-step loop before it answers, that combination matters more than a leaderboard win.
What Claude Sonnet 5.5 Actually Is
Sonnet 5.5 is the second model in Anthropic's Claude 5.5 family, released one week after Opus 5.5. It is not a new flagship. It is the tier most teams actually run in production, and this release is about making that tier good enough to stop reaching for the expensive one.
| Spec | Claude Sonnet 5.5 |
|---|---|
| Model ID | claude-sonnet-5-5 |
| Maker | Anthropic |
| Released | September 28, 2026 |
| Context window | 1,000,000 tokens |
| Max output | 128,000 tokens |
| Input | Text, image |
| Output | Text |
| Price | $2 per million input, $10 per million output |
| Cached input | $0.20 per million tokens |
| Knowledge cutoff | June 2026 |
Two rows carry the weight. A one-million-token context window means a research thread can hold a repository, a filing set, or a stack of source documents without a summarization pass in the middle. And cache reads at a fifth of the input price mean the repeated reads that a search loop depends on stop being the expensive part.
The price is unchanged from Sonnet 5. What changed is what you get for it.
The Numbers That Matter for Search
Anthropic published the launch benchmarks, and the pattern is worth reading carefully. Sonnet 5.5 does not win everything. It wins the columns that map to the work a search answer actually does.
| Benchmark | Sonnet 5.5 | Sonnet 5 | Opus 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 10.3% | 66.4% |
| FrontierCode 1.1 | 52.1% | 42.4% | 54.4% |
| CursorBench 4.0 | 55.5% | 34.1% | 57.8% |
| OSWorld 2.1 | 80.1% | 57.0% | 81.8% |
| Chartography | 61.6% | 15.6% | 64.4% |
Anthropic launch figures, September 2026. Terminal-Bench 4.0 and CursorBench 4.0 are reported at the model's best effort level; FrontierCode 1.1 is shown at xhigh.
Read the Sonnet 5 column and then the Sonnet 5.5 column. Terminal-Bench jumps from 10.3% to 70.6%. CursorBench nearly doubles. This is not a marginal refresh; it is a different model wearing the same price tag.
Where Opus 5.5 still leads, it leads by one to three points — and Anthropic's own guidance is that Opus remains the pick for complex, open-ended problems that need sustained judgment. For well-scoped everyday work, the gap is now small enough that the price decides.
Why This Changes What Felo Search Can Do
A single question in Felo Search is not one call to one model. It is a loop: decide what to look up, read what comes back, decide whether that was enough, look again, reconcile the sources, then write. Planning, tool selection, and retries all burn tokens, and a deep investigation can run that loop a dozen times.
That loop is where a model's price stops being an abstraction. At Opus pricing, a thorough multi-round investigation is a real line item, so the system has to be conservative about how many times it goes around. At Sonnet 5.5 pricing, the ceiling moves.
Two more things compound it:
- Output runs 30%+ faster than Sonnet 5. A search answer that takes several rounds feels different when each round returns sooner.
- It needs fewer tokens per task. Anthropic estimates up to 30% less to run on typical workloads. Balyasny Asset Management measured roughly 121,000 tokens per answer on 2,441 finance tasks, where Sonnet 5 used about 497,000. Box reported 2.4x faster output with 12% fewer total tokens. Slack saw better results in fewer steps and about 14% fewer output tokens.
Fewer tokens per answer, faster output, and a lower per-token price all point the same direction: more rounds of verification before the answer reaches you.
How to Use Claude Sonnet 5.5 in Felo Search
1. Open the model picker and select Claude Sonnet 5.5. Same search box, same sources, same workspace. Nothing else in the interface changes.
2. Ask questions that need more than one pass. Sonnet 5.5's advantage shows up on questions that require a loop: compare, verify, reconcile. A single fact lookup does not need this tier at all.
3. Use the full context instead of trimming. A million tokens is enough to keep every source attached to a long investigation rather than summarizing down to a page and losing the detail you will need later.
4. Let it check its own work. The jump on agentic benchmarks is the reason to trust a multi-step answer here. Ask for the caveats, not just the conclusion.
5. Escalate deliberately. When a question turns out to need sustained judgment, switch to a stronger model with the evidence already in context. The shared surface means nothing breaks in the handoff.
Where Sonnet 5.5 Fits in Felo Search
Felo Search runs several models, and the choice is about matching the tier to the question.
| Model | Price per million tokens | Context | Built for |
|---|---|---|---|
| Claude Sonnet 5.5 | $2 in / $10 out | 1M | Near-Opus quality on scoped, multi-step work |
| Claude Opus 5.5 | $4 in / $20 out | 1M | Answers that survive hard verification |
| GPT-6 Sol | $2 in / $10 out | 1.05M | Agent loops and repeat calls at mid-tier price |
| GPT-6 Luna | $0.10 in / $0.50 out | 1.05M | High-volume search at the lowest cost per call |
Sonnet 5.5 and GPT-6 Sol land at the same price. They differ in character: Sol carries a slightly larger context window, Sonnet 5.5 leads on agentic terminal coding and comes from a family Anthropic tunes for careful, well-scoped work. Run the same question on both and compare the caveats, not the prose.
What to Try First
A claim you intend to repeat. Pick a statistic you are about to put in a document and ask Sonnet 5.5 to trace it to its origin. The interesting output is the trail, not the number.
A three-way comparison. Ask for three vendors, their pricing, and the caveat each one buries. That question needs several retrieval rounds, which is exactly where the price per round matters.
A repository-scale question. Point it at a large body of documentation and ask how a specific behavior is implemented, with citations back to the source files.
A document you have to argue with. Paste a contract and ask what it does not cover. The context window is what makes "read all of it" a realistic instruction.
Try Claude Sonnet 5.5 in Felo Search
Claude Sonnet 5.5 is live in Felo Search with a one-million-token context window, image input, and $2/$10 pricing — the same price as Sonnet 5, with a benchmark jump that is anything but incremental. It beats Opus 5.5 on agentic terminal coding at half the cost, and it is available now.
Open Claude Sonnet 5.5 in Felo Search →
This post is also available in 简体中文, 日本語, 한국어, 繁體中文, हिन्दी, Français, العربية, Русский, اردو, Bahasa Indonesia, Deutsch, Tiếng Việt, Türkçe, Italiano, ไทย, Español, বাংলা and Português.