Will the next Claude Haiku model (4.6+) debut at a score of at least 1400 by June 30, 2027?

- [2026-10-07] Anthropic launched Claude Haiku 5.5 (`claude-haiku-5-5`), the first Haiku since Haiku 4.5. The name includes "Haiku" and the version is 5.5, so it should satisfy the market's "4.6+" test. Haiku 4.6 to 4.8 and Haiku 5 do not exist. Anthropic said Haiku 5.5 would follow Opus 5.5 (Sep 22) and Sonnet 5.5 (Sep 28). https://www.anthropic.com/ ; https://releasebot.io/updates/anthropic ; https://hidekazu-konishi.com/entry/anthropic_claude_model_release_timeline.html - [2026-10-07] Arena.ai posted on X that Haiku 5.5 is "now in the Arena". It is in Agent Arena, with "scores coming soon". It is also available for testing in Code Arena (WebDev), Text, Document and Vision. This is availability for voting, not a published Text score. https://x.com/arena/status/2107906881276826082 - [2026-10-13] The arena.ai homepage "New Release Rankings" box lists Haiku 5.5 as #30 in WebDev (High). Its snippet showed no Text ranking. I did not open the page itself. https://arena.ai/leaderboard - [2026-10-13] On the resolution page (header dated Oct 8, 2026; 414 models; top model 1534), I found no Haiku 5.5 entry in the rows shown (about the top 58). That is evidence of absence, not proof. The only Haiku on the page is claude-haiku-4-5-20251001, at rank 156 with 1396 (±2) and 146,351 votes. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-12] The Arena changelog, current through Oct 12, has no Haiku 5.5 entry for any leaderboard. The last Text Arena additions were claude-sonnet-5.5-xhigh (Oct 2), gemini-4-argon-high (Sep 30), mistral-large-4 (Oct 8) and others on Sep 25. So Haiku 5.5 had not been added to the Text leaderboard as of Oct 12. https://arena.ai/company/leaderboard-changelog - [2026-10-13] Comparable Anthropic and small-model scores on the same no-style-control board: Sonnet 5.5 (xhigh) 1471 (preliminary-sized, 4,838 votes); Sonnet 4.6 1458; Sonnet 5 (high) 1442; Opus 4.8 1454. Other small or fast models: gemini-3.5-flash-lite 1435; gpt-6-luna-max (a cheap OpenAI small model) 1391; gpt-5.4-mini-high 1411; gpt-5.6-luna-xhigh 1430. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-07] Anthropic says Haiku 5.5 is its most capable small model and beats Haiku 4.5 in coding, computer use and knowledge work. It costs about 75% less than Haiku 4.5, at $0.10/$0.50 per million tokens. It has a 1M-token context window. https://finance.yahoo.com/technology/article/anthropic-reveals-haiku-55-model-as-ai-pricing-war-intensifies-180000423.html ; https://benchlm.ai/models/claude-haiku-5-5 - [2026-10-08 to 10-13] Third-party benchmarks so far: Artificial Analysis Intelligence Index 43.4 (Max effort, via OpenRouter); LiveBench average 72.1, 42nd of 57; BenchLM 66.8. As of Oct 8, no independent lab had published results on LMArena. A review noted that the Arena human-preference score "usually" appears a few days after launch. https://openrouter.ai/anthropic/claude-haiku-5.5 ; https://roo3.co/en/articles/haiku-5-5-benchmarks-and-pricing ; https://apidog.com/blog/claude-haiku-5-5-benchmarks/ - [2025-10 precedent] Haiku 4.5 (released Oct 15, 2025) was added to the Text Arena within days of launch, per the changelog. It now sits at 1396. https://arena.ai/blog/leaderboard-changelog - [2026-10] Scheduling and risk: the Haiku 5.5 Text score is likely to appear within days or weeks. The market's "first appears" date and the 12:00 PM ET next-day score are what count. A model labeled "AutoEval" does not count. Only models added on the first appearance date compete, and Haiku 5.5 may be the only qualifying one. A further Haiku (e.g. 5.6) before the first appearance would not change much, because Haiku 5.5 already qualifies. https://polymarket.com/event/next-claude-haiku-released-byptptpt-20260701205353326 BOTTOM LINE: Haiku 5.5 launched Oct 7 and is already in Arena for voting. As of Oct 12–13 it had no published score on the Text overall (no style control) board, so the 1400 threshold will be tested against its first posted score soon. The benchmark is Haiku 4.5 at 1396, with newer Anthropic and small models mostly in the 1430–1470 range. NOT FOUND: A live Haiku 5.5 Text Arena score. The rendered leaderboard header was dated Oct 8, so a later refresh may exist. I also found no Arena statement of when the Text score will post, and no evidence of an "AutoEval" label for Haiku 5.5. I did not search for or report the market's own price or odds.

- [2026-10-07] Anthropic released Claude Haiku 5.5 (model ID claude-haiku-5-5). It is the first Haiku since 4.5 (2025-10-15), and Haiku 4.6, 4.7 and 4.8 never existed. By the contract's wording, "5.5" should qualify as a version above 4.6. https://www.anthropic.com/claude-haiku-5-5 and https://hidekazu-konishi.com/entry/anthropic_claude_model_release_timeline.html - [2026-10-07] Arena posted that Haiku 5.5 is "in the Arena" for Agent Arena, with Code/WebDev, Text, Document and Vision also available for testing, and "scores coming soon". Testing in Text battles is not the same as being on the public Text leaderboard. https://x.com/arena/status/2107906881276826082 - [2026-10-12] The Arena changelog, fetched in full, lists no Haiku 5.5 addition to the Text leaderboard. It shows claude-sonnet-5.5-xhigh added to Text on 10-02 and mistral-large-4 added to Text on 10-08. https://arena.ai/blog/leaderboard-changelog - [2026-10-08] The rendered no-style-control Text leaderboard (414 models) shows no Haiku in the top ~57 entries, all of which score 1447 or higher. Haiku 4.5 and 5.5 are not visible in the part of the page I could read. A first appearance is therefore probably still ahead, and a score of 1400 or more would put Haiku 5.5 near rank 70 or higher. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-12] Haiku 5.5's WebDev ranking appeared as "#30 in WebDev" on the Arena homepage. This shows Arena has published at least one score for it. https://arena.ai/leaderboard - [2025-10-15] Reference case: Haiku 4.5 was announced 2025-10-15 and added to Text Arena within days. The changelog lists it alongside amazon-nova-experimental-chat-10-09. Typical lag from release to Text listing is days to weeks. https://arena.ai/blog/leaderboard-changelog - [2026-10-08] Current scale: #1 is gemini-4-argon-high at 1534, and the Anthropic range is Opus 5.5-high 1515 down to claude-sonnet-5.5-xhigh 1471 (±9, 4,838 votes). Sonnet 4.6 scores 1458 and Opus 4.5 scores 1451. Google's cheap flash models score 1466–1499 (gemini-3.8-flash-high 1499, gemini-3.5-flash-high 1482). Small and cheap models now routinely score above 1400, e.g. glm-5.3-flash 1471 and mimo-v2.6-flash 1455. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-07] Haiku 5.5 benchmarks point to a mid-to-high score. It scores 43 on the Artificial Analysis Index at max effort, ahead of GLM-5.3 Flash (42) and Gemini 3.8 Flash (41). Vals Index is 54.31% (#16 of 45). That suggests it will land near other small frontier models, roughly 1450–1490 on this scale. This is my inference, not a published Arena figure. https://artificialanalysis.ai/articles/claude-haiku-5-5 and https://www.vals.ai/models/anthropic_claude-haiku-5-5 - [2026-03] Arena AutoEval: Arena introduced "AutoEval" provisional scores for new models, and the contract excludes models shown with that label. If Haiku 5.5 first appears as AutoEval and is later promoted, the "first appears" date and the score at the next-day noon check could be affected. https://arena.ai/blog/march-2026-arena-updates - [2026-02-20] Anthropic Text debut history: Sonnet 4.6 debuted at 1457 (#13) and Opus 4.6 at about 1502, both well above 1400. Models flagged "Preliminary" on the board, like gemini-4-argon-high, still carry a displayed score. https://x.com/arena/status/2024883614249615394 and https://www.codesota.com/arena/text - [2026-10-12] A third-party aggregator lists Claude Haiku 4.5's Arena score as "—". I could not confirm Haiku 4.5's debut score from a primary source (my recollection is roughly 1400–1410, unverified). https://anotherwrapper.com/tools/llm-pricing/evals/arena-text-elo BOTTOM LINE: The qualifying model very likely already exists (Haiku 5.5, released 2026-10-07 and in Arena testing) and is probably about to be added to the Text board. Current-era small and flash models score about 1450–1500, so 1400 is a low bar. The remaining risks are an AutoEval-only listing, a non-listing, or a title/timing dispute over whether 5.5 counts as the "next" model. NOT FOUND: Haiku 5.5's actual Text leaderboard score or listing date; Haiku 4.5's verified debut score from a primary source; any prediction-market or expert forecast for Haiku 5.5's Arena score (not searched, per instructions).

YES means: Anthropic's next Haiku model (version 4.6 or higher, most likely Claude Haiku 5.5) appears on the Arena Text overall (no style control) leaderboard without an AutoEval label, and its displayed Score at 12:00 PM ET the next day is 1400 or higher. This must happen by June 30, 2027. - [2026-10-12] The Arena changelog, fetched in full, is current through Oct 12. It has no Haiku 5.5 Text entry yet. Sonnet 5.5 took 4 days to reach Text (Sept 28 launch, added Oct 2), and Haiku 4.5 reached Text within days of its Oct 2025 launch. The listing is pending, not refused. Arena's Oct 7 post says Haiku 5.5 is open for Text battles, so votes are accumulating. https://arena.ai/company/leaderboard-changelog - [2026-10-13] The rendered Text board (dated Oct 8, 414 models) shows no new Haiku. It also shows how high the bar is not set. Cheap or flash-class models in the top 45 include glm-5.3-flash at 1471 ($0.06/$0.20), deepseek-v4.1-flash-max at 1462 and gemini-3-flash at 1466. mimo-v2.6-flash scores 1455 with a ±7 interval. Anthropic's lowest listed non-Haiku rows are Sonnet 4.6 at 1458 and Opus 4.5 at 1451. A Haiku at 1400 would sit roughly 55–70 points below its peers. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-10-13] The board shows "Preliminary" models (gemini-4-argon-high with 4,892 votes, gemini-3.8-flash-high) with plain numeric scores. Only "AutoEval" excludes a model under the contract. Arena's March 2026 post describes AutoEval as a stopgap while votes accumulate. Recent Anthropic additions (Sonnet 5.5 xhigh with 4,838 votes, Opus 5.5 high) show numeric scores with ±8–9 intervals, so a clean first listing is the normal pattern. https://arena.ai/blog/march-2026-arena-updates - [2026-10-13] Haiku 5.5's WebDev score is already published (#30, "High"). This is the first Arena leaderboard to carry it, and Text typically follows once enough votes accumulate. The score is already in the Arena pipeline. https://arena.ai/leaderboard - [2026-10-09] Third-party coverage (aiwiz.uk) confirms the LMArena Text listing was still missing on Oct 9. This matches the changelog and shows only a short delay. https://aiwiz.uk/blog/claude-haiku-5-5/ - [2026-10-08] The Artificial Analysis Intelligence Index gives Haiku 5.5 a score of 43 at max effort, ahead of GLM-5.3 Flash (42) and Gemini 3.8 Flash (41). Those two models score 1471 and 1499 on this board. Same-tier models on a comparable index already clear 1400 by 70 to 100 points. https://artificialanalysis.ai/articles/claude-haiku-5-5 - [2026-10-07] Haiku 5.5 improves hugely on Haiku 4.5. HLE without tools goes from 10.2% to 45.9%, and GDPval-AA Elo from 735 to 1620. Haiku 4.5, a much weaker model, still holds 1396 on the same board, only 4 points under the threshold. Haiku 5.5 only needs to match its predecessor's score. https://www.datacamp.com/blog/claude-haiku-5-5 - [2026-10-12] The Design Arena rating for Haiku 5.5 is 1355 and its code-category Elo is 1282. These are on a different scale and a different test, so they do not predict the Arena Text score. They show a human-vote-style Elo already exists for it from a third party. https://www.designarena.ai/models - [2026-10-07] Effort setting is a risk with an upside. Haiku 5.5 defaults to medium effort with adaptive thinking. Arena has listed multiple effort variants for other models (gemini-3.5-flash-high and -medium score 1482 and 1477). Medium or high Haiku variants could be added together, and under the contract the highest-scoring variant added that day counts. https://eesel.ai/blog/claude-haiku-5-5 (the page is eesel.ai/blog/claude-haiku-5-5; see https://www.eesel.ai/blog/claude-haiku-5-5) - [2026-10-13] There is time. The window runs to 2027-06-30, about 8.5 months. The only timing path to NO is the model not being listed, or being listed with an AutoEval label that is never removed, which is unlikely for a flagship-lab model that Arena already hosts. https://arena.ai/company/leaderboard-changelog Single most likely path to YES: Arena adds claude-haiku-5-5 (probably a high or max effort variant) to the Text board in the next days or weeks with a numeric score, with no AutoEval label. The score lands in the 1430–1490 range, in line with other small frontier models. It stays on the board through noon ET the next day. The only prerequisites are listing and a score at or above 1400. BOTTOM LINE: The model that qualifies already exists and is being collected in Arena, and its predecessor already scores 1396 on this board. Every comparable small or cheap model scores 55–100 points above 1400, so the score threshold is easy to clear if the listing happens cleanly. NOT FOUND: A live Haiku 5.5 Text score or listing date, any Arena statement on the Text posting schedule, and any evidence of an "AutoEval" label for Haiku 5.5. I did not find a verified debut score for Haiku 4.5.

YES means: the next Claude Haiku model (version 4.6 or higher; Haiku 5.5 launched 2026-10-07) appears on the Arena Text Overall no-style-control leaderboard without an AutoEval label, and its Score is at least 1400 at 12:00 PM ET on the next calendar day. - [2026-10-12] I fetched the full Arena changelog. It shows no Haiku 5.5 Text entry through Oct 12, five days after launch. Recent Anthropic Text additions came faster: Opus 5.5-high launched Sep 22 and reached Text on Sep 25, and Sonnet 5.5-xhigh launched Sep 28 and reached Text on Oct 2. Haiku 5.5 is already past that pace. A slip, a long wait, or no listing all favour NO. https://arena.ai/blog/leaderboard-changelog - [2026-10-12] The changelog shows Anthropic variants being added selectively. Sonnet 5.5-high went only to WebDev (Sep 29), Opus 5.5-max only to WebDev and Image-to-WebDev, and Sonnet 5.5 (Max) and Opus 5.5 (High) first to Agent Arena. Haiku 5.5 already has a WebDev score (#30, "High"), so Arena may route it to WebDev and Agent boards and skip Text. Arena's X post said only that Haiku was "available" in Text, with scores "coming soon". https://arena.ai/blog/leaderboard-changelog ; https://x.com/arena/status/2107906881276826082 - [2026-08 to 2026-10] Some 2026 releases did not reach the overall Text board promptly. A Aug 2026 explainer said Kimi K3 and Grok 4.5 were listed only on category boards, and the GPT-5.6 family "hasn't settled". Another explainer put the usual lag at 1–2 weeks. Arena has no commitment to list a small model on Text, and listing by 2027-06-30 is likely but not guaranteed. https://localaimaster.com/blog/lmarena-chatbot-arena-leaderboard - [2026-10-13] A same-price rival sits below the bar. gpt-6-luna-max, a max-effort small model priced at the same $0.10/$0.50, scores 1391 on this board (per the earlier report). Artificial Analysis rates Haiku 5.5 at 43 at max effort, 38 at high effort and 34 at the default medium effort. That 38 is level with Luna at max. If Arena lists a medium or high variant, or the score is dragged down by Haiku's verbosity, a result near 1400 is plausible. https://artificialanalysis.ai/articles/claude-haiku-5-5 ; https://beam.ai/agentic-insights/claude-haiku-5-5-subagents - [2026-10-13] The Haiku line is Anthropic's weakest on this board. Haiku 4.5 sits at 1396, below 1400, despite being a strong release in its day (per the earlier report). The older Haiku lineage also shows no Haiku score above 1400 on Text. Haiku 5.5's benchmark gains are mostly agentic (OSWorld, Terminal-Bench), and Arena Text measures chat preference. Higher agent scores need not translate into a higher chat Elo. https://anthropic.com/claude-haiku-5-5 ; https://myclaw.ai/blog/claude-haiku-5-5 - [2026-10-08] The only Haiku 5.5 crowd-preference number I found is WebDev 1587, from an aggregator (high effort). It sits 28th of the models ranked above it (Gemini 3.8 Flash scores 1580 at #28 on Code Arena), so it is unremarkable. No Text figure exists anywhere I checked. Any "1450+" claim is an inference from other benchmarks, not an Arena measurement. https://llm.ing/models/anthropic/claude-haiku-5-5 ; https://www.secondtalent.com/resources/gpt-6-vs-claude-vs-gemini/ - [2026-07-30] AutoEval is a live mechanism. Since Jul 30, Arena has shown Day-1 scores labelled "AutoEval" until enough human votes arrive. The contract ignores AutoEval rows. If Haiku 5.5 debuts as AutoEval, the "first appears" date could be pushed back or ambiguous. A preliminary, thin-vote score that is later revised downward also creates risk, since the check happens one day after first display. https://x.com/arena/status/2085380299371037017 - [2026-10-12] Resolution mechanics add NO paths. The market needs a clean listing that persists at noon ET the next day. It resolves NO if the model is removed before then, or if the page is unavailable for 7 days. It also resolves NO if no listing happens by 2027-06-30. The market can resolve only on the first qualifying listing. A score of 1396–1399 ("regardless of unrounded data") would fail narrowly. https://arena.ai/blog/leaderboard-changelog - [2026-10-13] Counter-context, which I am not weighting: Anthropic's mid-tier Text debuts have been well above 1400 (Sonnet 5.5 at 1471), and the board's top is about 1534. NO therefore depends on a low-probability tail: a non-listing, an AutoEval-only listing, or a Luna-like score under 1400. https://arena.ai/blog/leaderboard-changelog BOTTOM LINE: The strongest NO case is not that Haiku 5.5 is weak overall. It is that listing on the Text board is not guaranteed and is already slower than recent Anthropic launches. The first listing may be AutoEval, a variant, or a thin-vote score, and Haiku 4.5 and the same-price Luna both sit below 1400 on this board. NOT FOUND: A Haiku 5.5 score on the Text no-style-control board, and any Arena statement on when or whether it will be listed there. I also found no evidence of an AutoEval label on it, and no primary-source debut score for Haiku 4.5. I could not open Arena's X AutoEval article directly (access denied), so the AutoEval details come from the search snippet.