Ce pin suit l'actualité : il montre ce que l'on sait maintenant, et voici ce qui a changé et pourquoi.

1. Modifié12/10/2026 à 16:44par @ThePinGang

   Chapeau

   réécrit 31 ajoutés, 25 supprimés

   Anthropic's Anthropic releases Claude Haiku 5.5, the small, low-costthird model of theits Claude 5.5 family, is expected in mid-October 2026 after Anthropic said it would follow Opus 5.5 and Sonnet 5.5 "in the coming weeks".family and its cheapest, fastest small model, at $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, about 90% below Haiku 4.5.

   Début

   14 oct. 2026→7 oct. 20267 jours plus tôt

   Résumé

   réécrit 291 ajoutés, 193 supprimés

   Anthropic has promisedreleased Claude Haiku 5.5 twice: the Opus 5.5 announcement on 22 September said Sonnet 5.5 and Haiku 5.5 "will follow in the coming weeks, with manyon 7 October 2026, calling it "the cheapest, fastest, and most capable small model we've ever released"; it is the third model of the same improvements to performance, efficiency, and safety", and theClaude 5.5 family, after Opus 5.5 (22 September) and Sonnet 5.5 post (28 September), both of which had promised it "in the coming weeks" It is available "now on 28 September said Haiku 5.5 "will join all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure", and on the Claude 5.5 family in Platform as claude-haiku-5-5; it is the coming weeks" first Haiku-class model with an adjustable effort setting Anthropic describespositions it for high-volume, cost-sensitive work such as "built for high-volume summaries, compactions, database queries and cost-sensitive applications"; CNBC calls it "the cheapest offering in the suite", classification, and as a subagent alongside Opus 5.5 or Sonnet 5.5 on coding; it says Sonnet 5.5 and TechCrunch notes Anthropic "didn't give a firm date" The date Opus 5.5 remain better for complex agentic coding Notable features Price per million tokens for prompts up to 100K tokens is a prediction: last generation $0.10 input and $0.50 output (cache reads $0.01, cache writes $0.125), against Haiku 4.5 (15 October 2025) arrived 16 days after Sonnet 4.5 (29 September 2025), which from Sonnet 5.5's 28 September launch points 4.5's $1 and $5; above 100K tokens it is $0.50 input and $2.50 output. Anthropic puts that at 90% lower than Haiku 4.5 up to 100K tokens, 50% lower above, and about 14 October 2026 Until it ships, Anthropic's lineup still lists Claude Haiku 4.5 as 75% cheaper on average It uses an updated tokenizer, similar to Sonnet 5.5 and Opus 5.5, which takes slightly more tokens per task; Anthropic calls it its fastest model Notable features Anthropic has published no price, context window or benchmarks for to date at standard speed, though slower than Opus in Fast Mode Safeguards: Anthropic reports major alignment improvements over Haiku 5.5; it says only that it shares "many of the same improvements to performance, efficiency, 4.5; its cyber safeguards are stricter than Haiku 4.5's and a little looser than other recent models, and safety" as Opus 5.5 The rest of the Claude still block penetration testing, while its biology safeguards match Sonnet 5, Sonnet 5.5 family has a 1M-token and Opus 5 The page does not state a context window, 128K max output and a June 2026 knowledge cutoff, where Haiku 4.5 has 200K, 64K and February 2025 The model it replaces, Haiku 4.5, costs $1 per million input tokens and $5 per million output tokens, against Sonnet 5.5's $2 and $10window; the system card is at anthropic.com/claude-haiku-5-5-system-card Benchmarks BenchmarkHaiku 5.5Haiku 4.5GPT-6 LunaSonnet 5.5GDPval-AA v2.1 (knowledge work, Elo)162073514371840AA-Briefcase v1.1 (knowledge work)157861413361824OSWorld 2.1 (computer use, offline subset)72.4%15.7%48.9%83.9%Humanity's Last Exam (no tools)45.9%10.2%—56.9%Humanity's Last Exam (with tools)57.4%18.7%—64.5%Terminal-Bench 4.0 (agentic coding)39.2%0.0%16.4%70.6%FrontierCode 1.1 (Main) (agentic coding)46.4%—42.4%52.1% (Xhigh)Chartography (visual reasoning, no tools)46.4%6.4%29.1%61.6%