Will the next Sonnet model debut at a score of at least 1470 by December 31, 2026?

- [2026-09-28] Anthropic officially launched Claude Sonnet 5.5 as "the second model in our Claude 5.5 family," a faster, lower-cost complement to Opus 5.5, per Anthropic's release notes page. "We just launched Claude Sonnet 5.5, the second model in our Claude 5.5 family," described as "a faster, lower-cost complement to Claude Opus 5.5." This is the current "next Sonnet" candidate for the market. Source: support.claude.com/en/articles/12138966-release-notes - [2026-09-28] Anthropic's own Sonnet product page confirms the launch date and describes Sonnet 5.5 as "A clear upgrade over Sonnet 5 that runs 30% faster and costs up to 30% less for most work." Source: anthropic.com/claude/sonnet - [2026-09-29] VentureBeat reports independent benchmark comparisons for Sonnet 5.5 released alongside launch: the model reaches 80.1% on OSWorld 2.1 versus 81.8% for Opus 5.5 and 57% for Sonnet 5; on Chartography it scores 61.6% versus 64.4% for Opus 5.5 and 15.6% for Sonnet 5 , indicating a large jump over Sonnet 5 but still trailing Opus 5.5 on most tasks. Source: venturebeat.com - [2026-09-29 (age)] Artificial Analysis reports Sonnet 5.5 reaching #2 on its own Intelligence Index (not Arena), noting "Anthropic's new Sonnet model scores 56, just 2 points behind Opus 5.5 (max)" — this is a different metric than the Arena Elo-style Score used for resolution. - [2026-09-25 (leaderboard data date)] The resolution-source leaderboard itself (arena.ai/leaderboard/text/overall-no-style-control), fetched directly, shows Sonnet 5.5 is NOT yet listed. The highest-scoring Sonnet models currently on the board are claude-sonnet-4-6 at 1458, claude-sonnet-5-high at 1443, and claude-sonnet-4-5-20250929 at 1439 — all below the 1470 threshold. Source: arena.ai/leaderboard/text/overall-no-style-control - [2026-09-25] For context on the scale, the top overall model on the same leaderboard is claude-opus-5.5-high at 1518, and the 1470-line falls in a densely packed middle tier (e.g., glm-5.2-max at 1470, gpt-5.4-high at 1469, gpt-5.5-high at 1471), meaning 1470 is a real but not extreme bar. Source: arena.ai/leaderboard/text/overall-no-style-control - [2026-06-30] Claude Sonnet 5 launched; on its Arena debut/current listing it sits at 1443 (high variant), well under 1470, consistent with the pattern that no Sonnet-tier model has cleared 1460 on this leaderboard to date. Source: arena.ai/leaderboard/text/overall-no-style-control - [2026-02-17] Claude Sonnet 4.6 launched and is the highest-ever-scoring Sonnet model on the current board at 1458, still 12 points short of 1470. Source: anthropic.com/claude/sonnet; arena.ai leaderboard - [~2026-09-29] The Arena leaderboard changelog page (arena.ai/blog/leaderboard-changelog), the official log of additions, has visible entries only through August 19, 2026 in the fetched version and does not yet show a Sonnet 5.5 addition entry, indicating the model had not been confirmed added to the Text Arena leaderboard as of the fetch. Source: arena.ai/blog/leaderboard-changelog - [2026-09-22] Claude Opus 5.5 (the sibling flagship) was released shortly before Sonnet 5.5 and already occupies the #1 overall Arena slot at 1518, per third-party tracking of Anthropic's release cadence. Source: github.com/jqueryscript/anthropic-claude-timeline NOT FOUND: No direct arena.ai changelog entry or Arena screenshot confirming Claude Sonnet 5.5 has actually been added to the Text Arena "overall-no-style-control" leaderboard, nor its debut Score value in that specific table (as of the fetches performed, the live leaderboard snapshot obtained still reflected data only through Sep 25, 2026, i.e., pre-dating the Sep 28 Sonnet 5.5 launch). BOTTOM LINE: Claude Sonnet 5.5 launched Sep 28, 2026 and is the presumptive "next Sonnet" for this market, but as of the most recent fetch it had not yet appeared on the arena.ai Text Arena leaderboard; historically, no Sonnet-family model (4.5, 4.6, or 5) has ever scored above 1458 on this leaderboard, suggesting a debut at or above 1470 would break from Sonnet's established scoring pattern even though Sonnet 5.5 shows large benchmark gains over Sonnet 5 on other (non-Arena) evaluations.

- [2026-09-25] The live Arena.ai Text Arena (no style control) leaderboard shows Claude Sonnet 4.6 at 1458, Claude Sonnet 5 (high) at 1443, and Claude Sonnet 4.5 (20250929) at 1439 — all three of the last Sonnet releases currently sit below the 1470 threshold, while the top model (Claude Opus 5.5 high) sits at 1518 and only ~28 of 409 models clear 1470. https://arena.ai/leaderboard/text/overall-no-style-control - [2025-09-29] At its debut, Claude Sonnet 4.5 was reported to be "tied for #1 in the Text Arena, matching Claude Opus 4.1" — a strong relative-rank showing, but on the current, more-populated leaderboard its score has settled at 1439, below 1470, illustrating that early "top-tier" framing does not guarantee retaining a 1470+ score as more models/votes are added. https://x.com/arena/status/1973828836510085385 - [2026-02-20] Claude Sonnet 4.6 launched and "landed #3 in Code and #13 in Text Arena," a solid but not top-2/3 overall-text result; its current overall score (1458) sits under 1470. https://x.com/arena/status/2024892333452656898?lang=en - [2026-07-02] Claude Sonnet 5 (thinking) was added to the Text Arena per Arena's official changelog; it currently scores 1443, meaningfully below Opus-tier models and below the 1470 bar. https://arena.ai/blog/leaderboard-changelog - [2026-09-28] Anthropic released Claude Sonnet 5.5, explicitly pitched as a faster/cheaper "complement" to Opus 5.5 for "well-scoped everyday tasks," not as a frontier-pushing model — matching the pattern of Sonnet being a mid-tier, not top-of-arena, product. https://www.anthropic.com/claude-sonnet-5-5 - [2026-09-28] As of the latest Arena leaderboard changelog snapshot (through Sept 28), Sonnet 5.5 had not yet been added to the Text Arena leaderboard, so its qualifying debut score for this market is still pending. https://arena.ai/blog/leaderboard-changelog - [2026-02-26] A third-party benchmark roundup for late-Feb 2026 placed Anthropic's Opus models and Gemini/Grok at the very top of Text Arena (Opus 4.6 tied #1 at 1503, Gemini 3.1 Pro #3 at 1500, GPT-5.2 #6 at 1481), underscoring that 1470+ has generally required flagship (Opus-class or rival frontier) models rather than Sonnet-tier releases. https://www.buildmvpfast.com/blog/claude-opus-4-6-lmsys-arena-benchmark-comparison-2026 - [2026-09-25] Score inflation/decay dynamics: the current Text Arena spread runs from 1518 (top) down through the 1470s occupied largely by Opus variants, Gemini 3.x, GPT-5.5/5.4-high, and open-weight frontier models (GLM-5.3, Qwen3.5-max, Kimi-K3-max) — Sonnet-branded models have consistently clustered in the 1435–1460 band rather than the ≥1470 band. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-04-27] An analysis of the arena data described Anthropic's positioning explicitly as "Opus 4.6 is the strongest general-purpose model available right now, Sonnet 4.6 is the best value for most production use cases," consistent with Sonnet being deliberately tiered below the frontier score band. https://www.buildmvpfast.com/blog/claude-opus-4-6-lmsys-arena-benchmark-comparison-2026 BOTTOM LINE: Across the last three Sonnet-branded releases (4.5, 4.6, 5), none currently holds a Text Arena (no-style-control) score at or above 1470 — each has landed in the roughly 1435–1460 range, well below Opus-class and other frontier models that occupy the ≥1470 band; this base rate suggests a new Sonnet (5.5 or later) debuting at ≥1470 would be a historical outlier for the Sonnet line rather than the norm. NOT FOUND: An exact, contemporaneous "debut day" score for Claude Sonnet 4.5/4.6/5 at the moment they were first added (only current, possibly vote-adjusted, values were available); no confirmation yet of Sonnet 5.5's specific addition date/score on the Arena Text leaderboard, as it appears not yet listed as of the last changelog update (Sept 28, 2026).

YES means: Claude Sonnet 5.5 (or whichever "next Sonnet" model first appears on the arena.ai Text Arena "overall-no-style-control" leaderboard) debuts with a Score column value of at least 1470 as displayed at 12:00 PM ET the day after its first appearance. - [2026-09-25] The precise leaderboard band around the 1470 threshold is thin and crowded with non-flagship models, not just Opus-class systems: glm-5.2-max sits at exactly 1470, gpt-5.4-high at 1469, and glm-5.3-max/gpt-5.5-high cluster at 1471-1472, showing the bar is a mid-pack cutoff rather than an exclusively frontier-flagship line — a mid-tier "workhorse" model reaching 1470 is not unprecedented on this board. https://arena.ai/leaderboard/text/overall-no-style-control (fetched directly) - [2026-09-28] Anthropic's own launch materials describe Sonnet 5.5 in benchmark terms that differ sharply from prior Sonnet launches: on Terminal-Bench 4.0 it jumps to 70.6% from Sonnet 5's 10.3%, and on GDPval-AA it scores "two points below Opus 5.5," a far smaller gap to the flagship than any previous Sonnet generation showed against its contemporaneous Opus. https://www.anthropic.com/claude-sonnet-5-5 - [2026-09-29] Independent evaluator Artificial Analysis finds Sonnet 5.5 reaches #2 overall on its Intelligence Index, with the model gaining "18 points over Sonnet 5" and moving ahead of GPT-6 Astra and GPT-6 Sol into the runner-up slot behind only Opus 5.5 — a positioning no prior Sonnet has achieved relative to that period's Opus and OpenAI flagships. https://artificialanalysis.ai/articles/claude-sonnet-5-5 - [2026-09-29] The same Artificial Analysis review notes Sonnet 5.5 "reaches parity with Opus 5.5" on AA-Briefcase, GDPval-AA, and AutomationBench-AA specifically at high effort settings, framing this generation's gap-closing as a deliberate design outcome (achieved via much heavier token usage) rather than an incremental tweak. https://artificialanalysis.ai/articles/claude-sonnet-5-5 - [2026-09-29] Third-party outlet coverage independently frames Sonnet 5.5 as "A midrange model that beats OpenAI's best" on the Intelligence Index, explicitly crediting Anthropic with taking the top two overall slots on that ranking (Opus 5.5 and Sonnet 5.5), which if any of that relative strength transfers to human Arena voting preference, would support Sonnet 5.5 clearing the crowded ~1470 band. https://www.trendingtopics.eu/anthropic-sonnet-5-5-overtakes-gpt-6/ - [2026-09-28] Anthropic frames Sonnet 5.5 not as a discount model but as being "strongest at well-scoped everyday tasks, fixing bugs, and creating polished documents, slides, and spreadsheets" plus having "a sharp eye for design" — attributes (formatting, presentation, task polish) that historically correlate with strong human Arena preference votes independent of raw benchmark scores. https://www.anthropic.com/claude-sonnet-5-5 - [2026-09-25] Precedent exists for a Sonnet model briefly reaching the very top of the Text Arena: Claude Sonnet 4.5 was reported "tied for #1 in the Text Arena, matching Claude Opus 4.1" at its debut, establishing that Anthropic's Sonnet line has previously achieved top-tier Arena standing (even though its score later settled lower as more votes/models were added), which is the closest historical analog for a fast, favorable initial Arena placement. https://x.com/arena/status/1973828836510085385 - [2026-09-29] As of the latest fetch, arena.ai's Text Arena leaderboard snapshot is still dated Sep 25, 2026 and does not yet list Sonnet 5.5, meaning the debut score is not yet locked in and the highest-scoring current Sonnet variants (claude-sonnet-4-6 at 1458) still leave headroom of only 12 points to the 1470 bar — a gap that Sonnet 5.5's documented benchmark jumps (e.g., +50 points on Terminal-Bench 4.0, +18 on the AA Intelligence Index) plausibly could close given the scale of improvement over Sonnet 5. https://arena.ai/leaderboard/text/overall-no-style-control BOTTOM LINE: The strongest case for YES rests on Sonnet 5.5's unusually small benchmark gap to Opus 5.5 (closest of any Sonnet generation to its flagship sibling) combined with the historical precedent that Sonnet 4.5 briefly tied for #1 in Text Arena at debut, together suggesting Sonnet 5.5 could plausibly clear the thin, non-exclusively-flagship 1470 band once it is added to the leaderboard — though this remains unconfirmed since Sonnet 5.5 had not yet appeared on the arena.ai Text Arena leaderboard as of the last fetch. NOT FOUND: No confirmed arena.ai Text Arena listing, addition date, or debut Score for Claude Sonnet 5.5 exists yet in any source checked; no Arena.ai (official account) statement specifically addressing Sonnet 5.5's Text Arena performance was found.

YES means: the next Anthropic-branded "Sonnet" model (currently Claude Sonnet 5.5, released Sep 28, 2026, or any later Sonnet variant that debuts first) shows a Score ≥1470 in the Score column of arena.ai's Text Arena "Overall, no style control" table at 12:00 PM ET on the calendar day after its first appearance there, per the rules; this dimension argues for NO. - [2026-09-28] Arena.ai's own official changelog (fetched directly), which lists every addition to the Text Arena leaderboard day-by-day, shows entries through September 28, 2026 that include Opus 5.5, DeepSeek, Grok, GPT-6 and Gemini variants added to Text Arena on Sept 25 and Agent Arena additions on Sept 27–28 — but contains no line item adding "claude-sonnet-5-5" (or any Sonnet 5.5 variant) to the Text Arena leaderboard as of that date. https://arena.ai/company/leaderboard-changelog - [2026-09-25] By contrast, Claude Opus 5.5 (High) — released Sept 22 per third-party tracking — was already added to the Text Arena leaderboard on Sept 25, i.e., roughly 3 days after release; Sonnet 5.5 has now gone at least a day longer than that gap without a confirmed Text Arena listing, per the same changelog. https://arena.ai/company/leaderboard-changelog - [2026-07-30] Arena.ai's "AutoEval" system, introduced by Arena's own product blog, gives new models a "Day-1 signal" using a reward-model proxy for votes, and these scores are "clearly labeled 'AutoEval,' then updated once enough human votes arrive to validate them." https://arena.ai/blog/autoeval-scores - [2026-07-30] Per the same Arena blog post, AutoEval scores are described as highly correlated with eventual live scores but explicitly a placeholder mechanism run "in under an hour" that is later replaced — this is precisely the labeled-"AutoEval" state the market's own resolution rules say must NOT be used for resolution ("Only scores displayed without the 'AutoEval' label will be considered"), creating a live risk that Sonnet 5.5's very first appearance on the board is an AutoEval-tagged entry that doesn't yet count as "added." https://arena.ai/blog/autoeval-scores - [2026-09-25] The live leaderboard snapshot (409 models, 8,528,723 votes, dated Sept 25, 2026) confirms the top score on the entire board is 1518 (Opus 5.5-high) and that the 1470 line sits in a crowded mid-tier populated by GLM-5.2-max (1470), gpt-5.4-high (1469) and gpt-5.5-high (1471) — with every currently-listed Sonnet variant (claude-sonnet-4-6 at 1458 down to claude-sonnet-4-5 variants at 1435–1439) still clustered 12+ points under the bar. https://arena.ai/leaderboard/text/overall-no-style-control - [2026-09-29 (12h old)] Independent benchmark trackers (Artificial Analysis, via 24/7 Wall St.) frame Sonnet 5.5 as deliberately positioned as the cheaper, faster "complement" to Opus 5.5 rather than a frontier-score contender, consistent with Anthropic's own product-tier strategy of keeping Sonnet below Opus on capability metrics. https://247wallst.com/cards/xpost-01m3mp9t61dye3ytyggdh1a6zn - [2026-09-28] Anthropic's own release notes and product page repeatedly frame Sonnet 5.5 as "a faster, lower-cost complement" and a model "that runs 30% faster and costs up to 30% less," language emphasizing efficiency/value rather than benchmark-leading intelligence — the same positioning under which every prior Sonnet release (4.5, 4.6, 5) has landed below 1470 on this exact leaderboard. https://www.anthropic.com/claude/sonnet - [Resolution-criteria mechanism] The rules state that if a model is first added but later removed and not displayed at the T+1 12:00 PM ET check, "its appearance will not qualify as added," and that any subsequent higher-scoring Sonnet variant added even one calendar day later than the first qualifying model "will not be considered" — both clauses give the market multiple narrow, mechanical paths to NO even if a later Sonnet score eventually clears 1470, so long as the specific qualifying snapshot does not. BOTTOM LINE: As of the most recent Arena changelog fetch (through Sept 28, 2026), Sonnet 5.5 had still not been confirmed added to the Text Arena leaderboard, and Anthropic's own positioning plus the unbroken 4.5/4.6/5 precedent (all landing 1435–1458) means the two live risks for NO are (a) Sonnet 5.5 debuting with a genuine Score below 1470 as its predecessors did, and (b) the AutoEval-label/removal/next-day-cutoff technicalities in the rules preventing an otherwise-qualifying high score from counting at the exact T+1 checkpoint. NOT FOUND: No confirmed Arena Text-leaderboard row, changelog line, or debut Score for Claude Sonnet 5.5 as of the latest fetch (Sept 28, 2026 changelog); no evidence yet of whether its first appearance (whenever it occurs) will carry an "AutoEval" label at the T+1 12:00 PM ET resolution checkpoint.