Claude Sonnet 5.5 is the mid-tier Claude, released September 28, 2026 as a faster, lower-cost complement to Opus 5.5. It scores close to the flagship: 56 on the Artificial Analysis Intelligence Index against Opus 5.5's 58, and 63.6% in Artificial Analysis's Terminal-Bench 4.0 run, the highest of the models on this hub. On the tbench.ai leaderboard it is #2 at 61.8%, and it is #3 on LMArena's WebDev board. Anthropic reports 70.6% on Terminal-Bench 4.0 against 10.3% for Sonnet 5, and says it runs over 30% faster and costs up to 30% less than its predecessor. It has a 1M-token context window and 128K output; effort defaults to high on the API and medium in Claude Code.
Strengths
+ #2 on the tbench.ai Terminal-Bench 4.0 leaderboard (61.8%), behind Opus 5.5
+ Half the per-token price of Opus 5.5 ($2/$10 vs $4/$20)
+ About 136 output tokens/s at max effort, per Artificial Analysis
Trade-offs
− Artificial Analysis flags it as very verbose, which adds output tokens
− A cache miss bills input at $2 per million, 20 times the cache-read rate
− Released September 2026, so its long-run track record is still short
Claude Opus 5.5 is Anthropic's current flagship, released September 22, 2026. Anthropic says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, and its docs recommend it as the default starting model. It scores 58 on the Artificial Analysis Intelligence Index (max effort), the highest of the models on this hub. On Terminal-Bench 4.0 it leads the tbench.ai leaderboard at 64.9% in Claude Code; in Artificial Analysis's own Terminal-Bench 4.0 run it scores 59.6%, behind Sonnet 5.5 at 63.6%. It has a 1M-token context window, 128K output, and always-on adaptive thinking with medium effort by default.
Strengths
+ #1 on the tbench.ai Terminal-Bench 4.0 leaderboard (64.9% in Claude Code)
+ Highest Artificial Analysis Intelligence Index score on this hub (58)
+ 1M-token context at the standard rate, with 128K output
Trade-offs
− Costs twice Sonnet 5.5 per request ($0.107 vs $0.053) for a small measured gain
− Sonnet 5.5 scores higher in Artificial Analysis's Terminal-Bench 4.0 run (63.6% vs 59.6%)
− Released September 2026, so its long-run track record is still short
Claude Haiku 5.5 is Anthropic's fast, low-cost model, released October 7, 2026 and pitched for classification, extraction, routing and coding subagents under Opus 5.5 or Sonnet 5.5. It is the first Haiku with adjustable effort and a 1M-token context window. Artificial Analysis scores it 43 on its Intelligence Index (max effort), against 17 for Haiku 4.5, at about 238 output tokens per second, and 32.8% in its Terminal-Bench 4.0 run; Anthropic reports 39.2% on Terminal-Bench 4.0, where Haiku 4.5 scored 0%. Pricing has two bands: prompts over 100K tokens are billed at five times the base rate on every token, and 49.7% of the coding-agent requests we measured were that large.
Strengths
+ 1M-token context and adjustable effort, both new for Haiku
+ Scores 43 on the Artificial Analysis Intelligence Index, against 17 for Haiku 4.5
+ About 238 output tokens/s, per Artificial Analysis
Trade-offs
− Prompts over 100K tokens cost 5x on the whole request, and 49.7% of the coding-agent requests we measured were that large
− New tokenizer counts about 30% more tokens than Haiku 4.5 for the same text
Claude Fable 5.1 is Anthropic's premium model, released September 1, 2026 alongside the limited-access Mythos 5.1. Anthropic says it is about 25% cheaper than Fable 5 on typical work and up to 45% cheaper on agentic work, thanks to a lower cache-read price. It scores 53 on the Artificial Analysis Intelligence Index and 52.0% in Artificial Analysis's Terminal-Bench 4.0 run (57.9% on the tbench.ai leaderboard), and ranks #2 on Scale's SWE-Bench Pro V2. Three weeks after launch, Anthropic said Opus 5.5 performs at Fable 5.1's level on most work; Opus 5.5 scores higher on both boards and costs less than half as much per request ($0.107 vs $0.24).
Strengths
+ #2 on Scale's SWE-Bench Pro V2 (99.1, Claude Code harness)
+ Cache reads at $0.25 per million, a quarter of Fable 5's
+ 1M-token context with 128K output
Trade-offs
− Opus 5.5 scores higher on Artificial Analysis and Terminal-Bench 4.0 at less than half the cost per request
− About 71 output tokens/s at max effort, per Artificial Analysis
3,400+ agent users