Claude Sonnet 5 vs Claude Sonnet 5.5
Same $2/$10 pricing, completely different model — Sonnet 5.5 is 77% faster, scores +18 on the Intelligence Index, and jumps 60 points on Terminal-Bench.
Claude Sonnet 5 launched on June 30, 2026; Claude Sonnet 5.5 followed three months later on September 28. Both cost $2/$10 per MTok, but Sonnet 5.5 delivers a generational leap in agentic coding, computer use and visual recognition — while outputting tokens 77% faster. Artificial Analysis also measures a higher cost per task at max effort. This Claude Sonnet 5 vs Sonnet 5.5 comparison covers 9 benchmarks, speed, per-task cost, and whether there is any reason to stay on Sonnet 5.
Last updated 2026-10-01 · Vendor data sourced from Anthropic Sonnet 5 and Anthropic Sonnet 5.5
At a Glance: Claude Sonnet 5 or Claude Sonnet 5.5?
Stay on Claude Sonnet 5 only when…
You need to disable thinking mode entirely for non-reasoning requests, or you run cost-sensitive max-effort workloads you have not re-tested on 5.5. For everything else, Sonnet 5.5 is the better model.
Upgrade to Claude Sonnet 5.5 when…
Almost always. Same per-token price, 77% faster output, +47% Intelligence Index, +60 pp agentic coding and 5 months fresher knowledge. Re-test cost per task at max effort first.
Claude Sonnet 5 vs Sonnet 5.5: What Changed
Every difference between the two models — sourced from Anthropic and Artificial Analysis.
| Claude Sonnet 5 | Claude Sonnet 5.5 | |
|---|---|---|
| Intelligence Index (max) | 38 | 56 (+18) |
| Terminal-Bench 4.0 | 10.3% | 70.6% (+60.3 pp) |
| OSWorld 2.1 | 57.0% | 80.1% (+23.1 pp) |
| Chartography | 15.6% | 61.6% (+46.0 pp) |
| Output speed | ~79 t/s | ~139 t/s (77% faster) |
| Cost per II task (AA, max effort) | $5.09 | $7.62 (+50%) |
| Knowledge cutoff | Jan 2026 | Jun 2026 (+5 months) |
| Thinking mode | Adaptive (can be disabled) | Adaptive (always on) |
| Pricing | $2 / $10 per MTok | $2 / $10 per MTok (identical) |
Sonnet 5.5 improves quality and speed on every axis. The trade-offs: thinking mode can no longer be disabled, and cost per task at max effort is higher in Artificial Analysis testing. At the same $2/$10 pricing, it is still one of the most lopsided quality upgrades in the Claude model family.
Claude Sonnet 5 vs Sonnet 5.5: Key Specifications
API identifiers, pricing, context limits and reasoning behavior — sourced from Anthropic's official documentation.
| Spec | Claude Sonnet 5 | Claude Sonnet 5.5 |
|---|---|---|
| API model ID | claude-sonnet-5 | claude-sonnet-5-5 |
| Released | Jun 30, 2026 | Sep 28, 2026 |
| Tagline | Anthropic's balanced reasoning model (succeeded by Sonnet 5.5) | The best combination of speed and intelligence |
| Latency | Fast | Fast |
| Input price (per MTok) | $2 | $2 |
| Output price (per MTok) | $10 | $10 |
| Context window | 1M tokens | 1M tokens |
| Max output | 128K tokens (300K on Batch API) | 128K tokens (300K on Batch API) |
| Thinking | Adaptive; can be disabled | Adaptive; can be tuned with between_tools (no up-front thinking at effort <= high) |
| Default effort | high | high |
| Knowledge cutoff | Jan 2026 | Jun 2026 |
| Input modalities | Text + images | Text + images |
| Retirement | Not sooner than Jun 30, 2027 | Not sooner than Sep 28, 2027 |
Claude Sonnet 5 vs Sonnet 5.5: Identical Pricing, Different Cost per Task
Per-token pricing is identical at $2/$10 per MTok. Cache reads ($0.20), cache writes ($2.50 / $4.00), and batch pricing ($1/$5) are all the same. Neither model has fast mode.
| Pricing Tier | Claude Sonnet 5 | Claude Sonnet 5.5 |
|---|---|---|
| Standard API (per MTok) | $2 / $10 | $2 / $10 |
| Batch API (50% off) | $1 / $5 | $1 / $5 |
| Cache write (5 min) | $2.50 | $2.50 |
| Cache write (1 hour) | $4 | $4 |
| Cache read | $0.20 (0.1x base) | $0.20 (0.1x base) |
But per-task cost diverges. Artificial Analysis measured Sonnet 5 at $5.09 per Intelligence Index task and Sonnet 5.5 at $7.62, both at max effort (+50%), because Sonnet 5.5 generated more tokens (410M vs 370M). Anthropic's launch post says Sonnet 5.5 costs up to 30% less per task than Sonnet 5 and cites a finance customer using about 121k tokens per answer versus 497k. Savings depend on workload and effort level, so test on your own tasks.
Benchmark Results: Sonnet 5.5 Leads Every Evaluation
Official benchmarks from Anthropic's Sonnet 5.5 announcement and Artificial Analysis.
| Benchmark | Claude Sonnet 5 | Claude Sonnet 5.5 | Delta | Note |
|---|---|---|---|---|
| Terminal-Bench 4.0 | 10.3% | 70.6% | +60.3 pp | Agentic coding |
| Intelligence Index (max) | 38 | 56 | +18 | Artificial Analysis |
| CursorBench 4.0 | 34.1% | 55.5% | +21.4 pp | — |
| OSWorld 2.1 | 57.0% | 80.1% | +23.1 pp | Computer use |
| Chartography | 15.6% | 61.6% | +46.0 pp | Visual recognition |
| GDPval-AA v2.1 | 1449 Elo | 1844 Elo | +395 Elo | Knowledge work |
| AA-Briefcase v1.1 | 1359 Elo | 1811 Elo | +452 Elo | Business reasoning |
| Humanity's Last Exam | 54.9% | 64.5% | +9.6 pp | With tools |
| FrontierCode 1.1 Main | 42.4% | 46.2% | +3.8 pp | Max effort |
Terminal-Bench 4.0 shows the largest gain: +60.3 pp (10.3% to 70.6%). Sonnet 5 was barely functional for agentic coding; Sonnet 5.5 is a top-tier performer.
Visual recognition sees a similar leap: Chartography jumps from 15.6% to 61.6% (+46.0 pp). Sonnet 5.5 is the first Sonnet model to complete Pokemon Red working only from screenshots.
Business reasoning improves by 400+ Elo on both GDPval-AA (+395) and AA-Briefcase (+452), moving Sonnet 5.5 into the same tier as Opus 5.5 on knowledge work.
Output Speed: Claude Sonnet 5.5 Is 77% Faster
Sonnet 5.5 outputs approximately 139 tokens per second compared to Sonnet 5's ~79 tokens per second — a 77% speed improvement. This is unusual for a model upgrade: most successors trade speed for quality. Sonnet 5.5 improves both.
Raw speed is not the whole story. Artificial Analysis saw Sonnet 5.5 produce more total tokens than Sonnet 5 on its evaluation suite (410M vs 370M), so end-to-end time depends on how much the model reasons. Anthropic's customer data shows large token reductions on some workloads (about 121k vs 497k tokens per answer in finance), so measure latency on your own tasks.
Thinking Mode: The One Regression in Claude Sonnet 5.5
Sonnet 5 allows disabling thinking mode. You can send thinking: disabled to skip reasoning entirely — useful for trivial tasks where thinking adds latency and cost without improving results.
Sonnet 5.5 has always-on adaptive thinking. Sending thinking: disabled returns an error. The model always reasons adaptively, deciding how much to think based on the effort level. This is the only capability Sonnet 5 has that Sonnet 5.5 does not.
Both models default to high effort and support the same effort levels: low, medium, high, xhigh and max. The between_tools thinking mode (no up-front thinking at effort ≤ high) is available on Sonnet 5.5 as an additional steering option.
Claude Sonnet 5 vs Sonnet 5.5: Which to Choose by Use Case
Match your workload to the right model version.
| Scenario | Recommended | Why |
|---|---|---|
| Agentic coding | Sonnet 5.5 | Terminal-Bench +60.3 pp; CursorBench +21.4 pp |
| Computer use / automation | Sonnet 5.5 | OSWorld +23.1 pp (57% to 80.1%) |
| Visual recognition | Sonnet 5.5 | Chartography +46.0 pp (15.6% to 61.6%) |
| Business reasoning | Sonnet 5.5 | AA-Briefcase +452 Elo; GDPval-AA +395 Elo |
| High-throughput API calls | Sonnet 5.5 | 77% faster output (139 vs 79 t/s); test cost per task at your effort level |
| Knowledge-sensitive tasks | Sonnet 5.5 | Jun 2026 cutoff vs Jan 2026 — 5 months fresher |
| Non-reasoning trivial tasks | Sonnet 5 | Only model that allows thinking: disabled |
| New projects (greenfield) | Sonnet 5.5 | Strictly better at the same price |
Claude Sonnet 5 vs Sonnet 5.5: The Bottom Line
Claude Sonnet 5.5 is one of the most decisive upgrades in Anthropic's model history. At the same $2/$10 per MTok, it is 77% faster (~139 vs ~79 t/s), scores +18 on the Intelligence Index (38 to 56) and gains 60 points on Terminal-Bench (10.3% to 70.6%). Two trade-offs: thinking mode can no longer be disabled, and Artificial Analysis measures a higher cost per task at max effort ($7.62 vs $5.09), so re-test cost on your workload. Unless one of those is critical, there is no reason to stay on Claude Sonnet 5.
Claude Sonnet 5 vs Sonnet 5.5: FAQ
What are the main differences between Claude Sonnet 5 and Claude Sonnet 5.5?
Same $2/$10 per MTok pricing, but Sonnet 5.5 is dramatically better on quality and speed. Output speed jumps from ~79 to ~139 tokens per second (77% faster). Intelligence Index rises from 38 to 56 (+47%). Terminal-Bench goes from 10.3% to 70.6% (+60.3 pp). OSWorld improves from 57% to 80.1%. Knowledge cutoff extends from Jan 2026 to Jun 2026. The only regression: Sonnet 5.5 cannot disable thinking mode.
Is Claude Sonnet 5.5 faster than Claude Sonnet 5?
Yes, significantly. Sonnet 5.5 outputs roughly 139 tokens per second compared to Sonnet 5's ~79 tokens per second — about 77% faster. Raw speed is not the whole story: Artificial Analysis recorded more total output tokens for Sonnet 5.5 on its evaluation suite (410M vs 370M), so end-to-end time on long reasoning tasks depends on how much the model thinks. Measure latency on your own workload.
Does Claude Sonnet 5.5 cost the same as Claude Sonnet 5?
Per token, yes — both cost $2/$10 per million input/output tokens, with identical cache, batch and 1-hour cache write pricing. Per task it depends on workload and effort. Artificial Analysis measured $5.09 per Intelligence Index task for Sonnet 5 and $7.62 for Sonnet 5.5, both at max effort (+50%), because Sonnet 5.5 generated more tokens. Anthropic says Sonnet 5.5 costs up to 30% less per task than Sonnet 5 and cites a finance customer using about 121k tokens per answer versus 497k. Both can hold on different workloads, so test at your own effort level.
Should I upgrade from Claude Sonnet 5 to Claude Sonnet 5.5?
Yes for most workloads. Sonnet 5.5 scores higher on every benchmark, is 77% faster on output and has a 5-month newer knowledge cutoff, all at the same per-token price. Two reasons to hold off: you need to disable thinking mode (Sonnet 5.5 does not allow it), or you run cost-sensitive max-effort workloads, where Artificial Analysis measured a higher cost per task for Sonnet 5.5 ($7.62 vs $5.09). Re-test cost before switching.
How much better is Claude Sonnet 5.5 at coding than Claude Sonnet 5?
The gap is enormous on agentic coding. Terminal-Bench 4.0 jumps from 10.3% to 70.6% — a 60.3 percentage point improvement. CursorBench 4.0 goes from 34.1% to 55.5% (+21.4 pp). FrontierCode shows a smaller but still meaningful gain (+3.8 pp). Sonnet 5.5 is a generational leap for coding workflows.
Can Claude Sonnet 5.5 disable thinking mode?
No. Sonnet 5.5 has always-on adaptive thinking — sending thinking: disabled returns an error. Sonnet 5 allows disabling thinking mode. If your workflow requires non-reasoning responses (for lower latency or cost on trivial tasks), Sonnet 5 or a non-reasoning model is needed.
How does Claude Sonnet 5.5 compare to Claude Opus 5.5?
Sonnet 5.5 nearly matches Opus 5.5 on many benchmarks while costing half as much per token ($2/$10 vs $4/$20). On the Intelligence Index, Opus leads with 58 vs 56 for Sonnet 5.5. For most workloads, Sonnet 5.5 offers the better price-performance ratio.
What is Claude Sonnet 5's knowledge cutoff?
Claude Sonnet 5 has a knowledge cutoff of January 2026. Claude Sonnet 5.5 extends this to June 2026 — a 5-month advantage. For tasks requiring recent information, Sonnet 5.5 is the better choice.
Is Claude Sonnet 5 being deprecated?
Not immediately. Anthropic guarantees Sonnet 5 availability until at least June 30, 2027. However, Anthropic recommends migrating to Sonnet 5.5, which supersedes it at identical pricing with better performance across the board.
When were Claude Sonnet 5 and Claude Sonnet 5.5 released?
Claude Sonnet 5 launched on June 30, 2026. Claude Sonnet 5.5 followed on September 28, 2026 — about three months later. Both are part of the Claude 5 model family.
Related Comparisons
- Claude Opus 5.5 vs Claude Sonnet 5.5 — Anthropic's flagship vs mid-tier ($4/$20 vs $2/$10)
- Claude Fable 5.1 vs Claude Opus 5.5 — Anthropic's premium tier vs flagship
- Claude Opus 5.5 vs GPT-6 Sol — cross-vendor at the $2-$4/$10-$20 tier
- Claude Opus 5.5 vs Claude Sonnet 5 — flagship vs predecessor ($4/$20 vs $2/$10)
Sources & References
- Anthropic Claude Sonnet 5 Model Page
- Anthropic Claude Sonnet 5.5 Model Page
- Claude Sonnet 5.5 Announcement
- Anthropic Pricing
- Artificial Analysis: Claude Sonnet 5.5
- Artificial Analysis: Claude Sonnet 5
- Labellerr: Claude Sonnet 5.5 vs Sonnet 5
- BazaarLink: Claude Sonnet 5 vs 5.5 — Worth Switching?
- Kingy: Claude Sonnet 5.5 Specs, Benchmarks & Pricing