Will the next Claude Haiku model (4.6+) debut at a score of at least 1420 by June 30, 2027?

- [2026-10-07] Anthropic launched Claude Haiku 5.5 (model ID claude-haiku-5-5). It is the first new Haiku since 4.5 and the first Haiku above 4.5. It would qualify under the "whole-number generation" and "4.6 or higher" wording. Source: https://www.anthropic.com/claude-haiku-5-5 - [2026-10-07] Arena posted on X that Haiku 5.5 is in Agent Arena, with "scores coming soon". It is also available in Code Arena (WebDev, Text, Document, Vision). That is availability for voting, not a published Text score. Source: https://x.com/arena/status/2107906881276826082 - [2026-10-14] The Arena changelog says Claude Haiku 5.5 (High) was added to the Agent Arena leaderboard. It does not list Haiku 5.5 under Text Arena through Oct 15. Source: https://arena.ai/company/leaderboard-changelog - [2026-10-15] I rendered the Text Arena overall (style control off) page, dated Oct 15 with 416 models. The visible top ~57 rows show no Haiku 5.5. The middle of the page was truncated, so a lower row is not ruled out. No Haiku model appeared in what I could see. Source: https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Scores on that page for context: Gemini 4 Argon 1532, Claude Opus 5.5 (High) 1516, claude-sonnet-5.5-xhigh 1470, claude-sonnet-4-6 1458, claude-opus-4-8 1454. The 1420 threshold is about 50 points below Sonnet 5.5 and 38 below Sonnet 4.6. Source: https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-02] Sonnet 5.5 (released 09-28) was added to the Text Arena on 10-02, four days after release. Haiku 5.5, released 10-07, had no Text entry by 10-15. Source: https://arena.ai/company/leaderboard-changelog - [2026-10-14] Reports say Haiku 5.5 (High) ranks #27 in Agent Arena and #30 in WebDev. Source: https://cryptobriefing.com/claude-haiku-5-5-agent-arena-ranking/ - [2026-10-07] Third-party capability signals: Artificial Analysis Intelligence Index 43 for Haiku 5.5 versus 17 for Haiku 4.5. Anthropic's system card reportedly says it hallucinated about as much as Haiku 4.5. Source: https://aiwiz.uk/blog/claude-haiku-5-5 - [2026-10-15] Text Arena reference points for small and fast models: gemini-3.5-flash-high 1482, gemini-3-flash 1466, glm-5.3-flash 1471, mimo-v2.6-flash 1455. Haiku 4.5's current Text score was not visible in the portion I could read. Source: https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Other trackers (llmgateway, benchlm, GitHub timeline) confirm the Oct 7 release and name Haiku 5.5 as the latest Claude model. The Polymarket "released by" market lists Haiku 5.5 as the catalyst. Sources: https://llmgateway.io/timeline ; https://github.com/jqueryscript/anthropic-claude-timeline BOTTOM LINE: The qualifying model, Haiku 5.5, is already released but had no Text Arena (style-control-off) score as of Oct 15. If it is added by June 30, 2027, the question turns on whether its first displayed score reaches 1420. Peer small and fast models sit around 1455–1482, and Sonnet 5.5 is at 1470. One ambiguity is whether a later Haiku model would count as "next" if Haiku 5.5 is added first. NOT FOUND: Any Haiku 5.5 (or Haiku 4.5) row or score on the Text Arena overall page, since the middle of the rendered page was truncated. Also any Arena statement of when Haiku 5.5 Text scores will post, and any announced Haiku release after 5.5.

- [2026-10-07] Anthropic released Claude Haiku 5.5 (API id claude-haiku-5-5). This is the Haiku that most likely qualifies as "next (4.6+)", since 4.6–4.8 never existed. It is the first new Haiku in about a year, and Anthropic priced it at $0.10/$0.50 per million tokens. https://thenewstack.io/anthropic-claude-haiku-5-5/ and https://hidekazu-konishi.com/entry/anthropic_claude_model_release_timeline.html - [2026-10-15] The Arena text overall (style control off) page, dated Oct 15, lists 416 models. In the top ~211 rows that I read, Haiku 5.5 is not there. It therefore does not appear to be on this board yet. The arena.ai home page does show "Claude Haiku 5.5 is #27 in Agent · High", but that is the Agent board, which does not count. https://arena.ai/leaderboard/text/overall-no-style-control and https://arena.ai/leaderboard - [2026-10-15] Haiku 4.5, the predecessor, sits at 1397 (±2) on that board with 149k votes, rank ~159. The 1420 strike is 23 points above it. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Within-family gaps on the same board, from the pages I fetched: - Sonnet 4.5 is 1438, which is 41 above Haiku 4.5. - Sonnet 4.6 is 1458. - The newest Sonnet, claude-sonnet-5.5-xhigh, is 1470 (±8, 6.8k votes). - Opus 5.5 (High) is 1516. - If the Haiku–Sonnet gap stays around 40, Haiku 5.5 would land near 1430. That is my own extrapolation, not a source. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Small and cheap models from other labs on the same board: - gemini-3.5-flash-lite: 1435 - gpt-5.4-mini-high: 1411 - gpt-6-luna-max: 1392 - gpt-5.4-nano-high: 1372 - gemini-3.5-flash-high: 1482 - glm-5.3-flash: 1471 - Small-model scores range from about 1370 to 1480, so 1420 is near the middle of that range. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Score inflation: the top score is 1532 (Gemini 4 Argon, preliminary). Opus 4.6 is 1502 and Opus 5.5 is 1516. Newer Anthropic models are trending upward. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-08] As of Oct 8, no independent lab (including LMArena) had published Haiku 5.5 results. Anthropic's own benchmarks show large gains over Haiku 4.5 on agentic and computer-use tests: OSWorld 72.4% vs 15.7%, and Terminal-Bench 39.2% vs 0.0%. These are not chat-preference measures. https://apidog.com/blog/claude-haiku-5-5-benchmarks/ and https://www.marktechpost.com/2026/10/07/anthropic-releases-claude-haiku-5-5-a-small-model-with-1m-context-priced-at-0-10-per-million-input-tokens/ - [2025-10-15] Timing reference: Haiku 4.5 launched Oct 15, 2025. Its Arena text listing now shows 149k votes, but I did not find the date it first appeared. The Anthropic page shows Haiku 3.5 launched Oct 22, 2024. https://www.anthropic.com/claude/haiku?via=fahim - [2026-10-05] Sibling Polymarket "released by" market, per its page snapshot: December 31 was at 99% and October 31 at 89%. Those prices were stale, because Haiku 5.5 shipped Oct 7. That market covers the release, not the Arena score. https://polymarket.com/event/next-claude-haiku-released-byptptpt-20260701205353326 - [2026-10-16] Timing risk: Arena usually lists new models within days to weeks of release, and Haiku 5.5 is already about 9 days past launch without a text listing. If the listing only appears with "Preliminary" or AutoEval status, the market rules say it does not count as added. If Haiku 5.5 is added and scores below 1420, the market resolves No. A later Haiku would not count, because the first qualifying listing is what resolves the market. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Arena scores can shift when a model is first posted. One report says Arena re-baselined scores on Jul 12, 2026. https://localaimaster.com/blog/lmarena-chatbot-arena-leaderboard BOTTOM LINE: Haiku 5.5 has been released but is not yet on the Arena text board. The base rate is a first listing within weeks, almost certainly before June 2027. Haiku 4.5 sat at 1397, and the same-family gaps and small-model peers put plausible Haiku 5.5 scores around 1400–1450. 1420 is therefore near the median of that range, and a modest lean toward Yes is reasonable if it is added. NOT FOUND: Haiku 5.5's Arena text score or listing date. A published Arena or expert forecast of its score. The date Haiku 4.5 first appeared on the Arena text board. A Polymarket price for the 1420 strike or sibling strikes.

YES means: Claude Haiku 5.5 (or another Haiku 4.6+ model) is first listed on the Arena Text overall (style control off) leaderboard without an AutoEval label, and its displayed Score at 12:00 PM ET the next day is at least 1420, all by June 30, 2027. - [2026-10-07] Arena says Haiku 5.5 is live for voting in Code Arena (WebDev, Text, Document, Vision) as well as Agent Arena. A Text-arena battle pool therefore already exists, and a score is only waiting on enough votes. Latent.Space's Oct 7–8 recap also lists the Text arena as "scores pending." https://x.com/arena/status/2107906881276826082 - [2026-10-14] Haiku 5.5 (High) has been added to the Agent Arena leaderboard and is #27 there, with #30 in WebDev. That shows Arena publishes Haiku 5.5 results within a week for the other arenas, and Text follows the same pipeline. Arena's own changelog confirms the Agent Arena addition. https://arena.ai/company/leaderboard-changelog - [2026-10-08] Arena's changelog shows Text Arena additions can lag the Agent Arena release by 1–2 weeks for non-flagship models. Mistral Large 4 was added to Text on Oct 8, one day after Code Arena availability. Muse-spark-1.3-max was added to Text on Oct 13. Haiku 5.5's Text listing is plausibly imminent. https://arena.ai/company/leaderboard-changelog - [2026-10-07] The Artificial Analysis Intelligence Index, which the AA article places at 43 at max effort, puts Haiku 5.5 slightly ahead of GLM-5.3 Flash (42), Gemini 3.8 Flash (41) and GPT-6 Luna (38). Per the Oct 15 reports, glm-5.3-flash is 1471 and gpt-6-luna-max is 1392 on the Text board. Haiku 5.5 is therefore at least at peer level. The weakest peer (Luna) is the one that falls below 1420. https://artificialanalysis.ai/articles/claude-haiku-5-5 - [2026-10-07] Haiku 5.5 reaches 1578 Elo on AA-Briefcase, within the confidence intervals of GPT-6 Astra and Claude Fable 5.1. That is a human-style knowledge-work Elo, not an Arena chat score. It shows quality at the level of the large models on realistic tasks. https://artificialanalysis.ai/articles/claude-haiku-5-5 - [2026-10-07] Haiku 5.5 is the first Haiku with adaptive thinking and effort settings, and Arena lists it as "(High)". Reasoning-mode variants have typically scored higher on Text Arena than non-thinking small models. Haiku 4.5's 1397 predates this. Haiku 5.5 scoring 23+ points above it is a modest ask given the jump in capability. https://www.latent.space/p/ainews-claude-haiku-55-better-than - [2026-10-15] Haiku 5.5 is up against a Sonnet 5.5 that was added at 1470, and a family that has been rising. Opus 5.5 was at 1509 when Arena announced it #1 (BenchLeader/Arena news), up 18 over Opus 5. Haiku sitting 50 points below Sonnet 5.5 would still clear 1420. https://didcodexreset.com/news/93f8abd8a76a4fb0b9093b43.html - [2026-10-15] Reasons for caution that survive the YES case: AA reports a safety over-refusal issue seen in pre-release testing. Verbosity is high (about 162k output tokens per task at max). Over-refusal could cost preference votes. That is the main downside path to a score under 1420. https://artificialanalysis.ai/articles/claude-haiku-5-5 Most likely path to YES: Arena adds Haiku 5.5 (High) to the Text overall board as a normal, non-preliminary row within weeks. Its first-day score lands between peers (1435–1480) and its predecessor (1397), plus the roughly 40-point Haiku-to-Sonnet gap. The prerequisites are (a) non-AutoEval status, (b) the row being present at 12:00 PM ET the next day, and (c) a score of 1420 or more. Any 5.x Haiku released and listed before June 2027 also provides a second attempt only if Haiku 5.5 is never listed. BOTTOM LINE: Haiku 5.5 is already released, in the Text battle pool, and on the Agent and WebDev boards. A listing before June 2027 is very likely, so the question mostly rests on the score. Capability evidence (AA 43, ahead of GLM-5.3 Flash, which scores 1471 on Arena) puts a 1420+ debut within the plausible range. NOT FOUND: Haiku 5.5's Text Arena (style control off) row or score. Any Arena statement of the Text listing date. Any pre-listing leak or preliminary Text score. Haiku 4.5's first-listing date and debut score.

YES means: Anthropic's next Haiku (4.6 or higher, in practice Haiku 5.5) first appears on the Arena Text overall (style control off) board without an AutoEval label, and its Score at 12:00 PM ET the next day is 1420 or higher. - [2026-10-15] Arena's changelog, fetched in full for Sept 27 to Oct 15, lists Haiku 5.5 only under Agent Arena, on Oct 14. It has no Text Arena entry for Haiku 5.5. By Oct 15 Haiku 5.5 had been out 8 days with no listing, versus 4 days for Sonnet 5.5. Arena tweeted "scores coming soon" on Oct 7 for Agent Arena only. https://arena.ai/company/leaderboard-changelog - [2026-10-07] Arena's Oct 7 post says Haiku 5.5 is "also available in Code Arena," including a Text mode. Votes can therefore be collected, but nothing says when or whether it will be promoted to the main Text board. Latent.Space's recap says the Text score is "pending". https://www.latent.space/p/ainews-claude-haiku-55-better-than - [2026-10-15] The Text overall page lists 416 models. The visible top 57 rows run from 1532 down to 1448. There are 5 Gemini/Meta/Z.ai-class small or fast models above 1470, and Sonnet 5.5 sits at 1470 with only 6,831 votes. The 1420 threshold falls below row 57, in a region that is crowded with models. Haiku 4.5 (1397) and nothing Haiku-branded appears in the top rows. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Several recent entries carry the "Preliminary" tag, for example Gemini 4 Argon, Gemini 3.8 Flash and Gemini 3.7 Flash. This is a separate label from AutoEval. The rules only exclude AutoEval, so a Preliminary listing would count. A new entry's first score can also differ from the settled score, and the first-displayed score at +1 day is what resolves the market. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Haiku 5.5's gains are in agentic and knowledge-work areas. AA-Briefcase is 1578 Elo, and the Intelligence Index is 43, ahead of GLM-5.3 Flash at 42. Arena's Agent Arena rank is #27, with a +1.48% net improvement on 20.5K sessions. Agent rank is a weak guide to chat preference. https://x.com/arena - [2026-10-07] Artificial Analysis says the model uses far more tokens than GPT-6 Luna (about 55k vs 50k at equal score, and about 162k at max effort). It notes a safety over-refusal issue that Anthropic is fixing. Over-refusal and verbosity are known drags on head-to-head chat votes. The Text Arena entry is likely "High" effort, not "Max". https://artificialanalysis.ai/articles/claude-haiku-5-5 - [2026-10-15] Comparable small models: gpt-6-luna-max at 1392 is below 1420, despite being a 2026 flagship-generation small model. Haiku 4.5 at 1397 and GPT-5.4-mini at 1411 also sit below it. The Anthropic small-model tier has not cleared 1420 on this board before. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-15] Design Arena (a separate site) rates Haiku 5.5 at 1349 Elo, with its own scale. It is not comparable to Arena's scale. https://www.designarena.ai/models - [2026-10-16] Timing risk for a No: if Arena adds Haiku 5.5 only as an agent or code model, or removes it, or never lists a non-AutoEval Text entry before 2027-06-30, the market resolves No. Arena also sometimes skips Text for models it judges less relevant. Search found no Arena statement that a Text score is coming for Haiku 5.5. https://didcodexreset.com/news/93f8abd8a76a4fb0b9093b43.html BOTTOM LINE: The No case rests on three points. Haiku 5.5 has no non-AutoEval Text listing 9 days after launch. Anthropic's and others' small models have landed below 1420, with Haiku 4.5 at 1397 and Luna at 1392. Haiku 5.5's gains are agentic, with verbosity and over-refusal issues that hurt chat votes. Sonnet-tier peers cluster in the 1450s–1470s, but a Haiku debut of 1410–1419 would still resolve No. NOT FOUND: A Haiku 5.5 or Haiku 4.5 row, or any score, on the Text overall board. The date Haiku 4.5 first appeared on the Text board. Any Arena statement about when Haiku 5.5 Text scores will be published. The middle rows of the Text page were truncated in my render, so a lower-ranked Haiku 5.5 row cannot be ruled out.