The board · as of 1 Nov 2026

The picks

One current, defensible answer per category — with the reasoning, the challenger, and how long each title has been held. Click any pick for its full history and criteria.

Language

4 titles

Best overall LLM

Claude Fable 5.1

Joint first on the re-based Artificial Analysis Intelligence Index v4.3 (53, level with GPT-6 Astra), first on the official Terminal-Bench v4.0 board at 57.9%, and first on Vellum's quality/speed/cost composite at 65%. On Arena text it sits at 1504 inside a 1-6 tie band with the rest of the Anthropic top group. Costs more per Index task than GPT-6 Astra ($7.63 against $3.26), so price the alternative if volume matters.

Challenger: Gemini 4 Carbon (High) — Tops the live Artificial Analysis Intelligence Index v5 on 59, ahead of Claude Opus 5.5 (max) on 58 and GPT-6 Astra (max) on 57, with the incumbent down at 50 and outside the top eight. It is no longer a single-board lead: Arena's published 28 October data has Carbon first on text and second behind Claude Opus 5.5 on the overall agent board, though Vellum has not refreshed since 28 September. Proprietary and hosted, as the incumbent is, so no licence or availability constraint changes; note it costs $7.76 per Index task.

Artificial Analysis has re-based again: the live board is Intelligence Index v5, where Gemini 4 Carbon (high) leads on 59, Claude Opus 5.5 (max) is on 58 and GPT-6 Astra (max) on 57, while Claude Fable 5.1 reads 50 (xhigh, $10.85 per task) and 49 (max, $12.99) — so the v4.3 score of 53 and the $7.63 against $3.26 comparison no longer apply. On Terminal-Bench 4.0 the 57.9% figure is a Claude Code max-effort row on the official board, which runs each vendor in its own agent and so cannot rank vendors against each other; Vals AI's neutral mini-SWE-agent run puts Claude Opus 5.5 first at 65.15%, with Anthropic quoting 66.4% itself. Arena has moved on too: its published 28 October board has Gemini 4 Carbon and Claude Opus 5.5 at the top with Fable 5.1 fourth overall, so ignore the 2 September 1507/1504 snapshot and read the live board.

days of reign

since 10 Sep 2026

Best LLM for coding

GPT-6 Astra

GPT-6 Astra tops Artificial Analysis's independent Terminal-Bench v4.0 run at 59.6% on xhigh and 59.1% on max, with Claude Fable 5.1 next at 55.1% and Opus 5 absent from the published top three. Vals.ai has it first on Terminal-Bench 2.1 too, at 87.3% ahead of GPT-5.6 Sol (85.8%), Fable 5.1 (85.0%) and Opus 5 (84.6%). It has no independent SWE-bench Verified run — Vals archived that board as saturated — and it lists at $10/$50 per million tokens, 2.5x GPT-5.6 Sol, so price the high-volume repo-scale batches carefully.

Challenger: Claude Opus 5 — Claude Opus 5 still tops Vals' SWE-bench Verified board at 97.00%, but Vals archived that board as saturated in its 1 September 2026 update and never ran GPT-6 Astra on it. On Vals' Terminal-Bench 2.1, itself now archived in favour of 4.0, Opus 5 sits fifth at 84.64%, or 81.27% with its server-side Opus 4.8 fallbacks counted as failures, behind Opus 5.5, Astra, GPT-5.6 Sol and Fable 5.1. On Vals' Terminal-Bench 4.0 it manages 53.53% against Opus 5.5's 61.62%, so treat it as a legacy option rather than a live contender.

Vals' Terminal-Bench 2.1 table is now archived and superseded by Terminal-Bench 4.0 in its coding index: Claude Opus 5.5 leads at 87.64%, GPT-6 Astra is second at 87.27% and Opus 5 has slipped to fifth at 84.64%, or 81.27% once its server-side fallbacks to Opus 4.8 on refused tasks are counted as failures. Vals also archived SWE-bench Verified as saturated in its 1 September 2026 update, where Opus 5 finished top at 97.00% at $1.29 a test and the open-weight DeepSeek V4 Pro 0813 second at 96.40% for $0.10 a test. Lean on the Terminal-Bench 4.0 and repo-scale runs rather than either archived board.

days of reign

since 9 Sep 2026

Best open-weight LLM

GLM-5.3

GLM-5.3 (max) tops Artificial Analysis's open-weights table on Intelligence Index v4.3 with 45, one point clear of Kimi K3 (max) on 44, and the weights have been on Hugging Face since 28 August. On cost the two are effectively level — about $2.01 per Index task against K3's $2.00 — so the separator is throughput, with GLM-5.3 measured at roughly 66 output tokens per second against K3's 37. It ships under Z.ai's own glm-5.3 licence rather than a standard open one, and it takes text-only inputs.

Challenger: DeepSeek-V4.1-Flash-1017 — MIT-licensed, weights out and widely served in vLLM and SGLang. Artificial Analysis has now benchmarked it independently: 32 on Intelligence Index v5.0, level with GLM-5.3 (max) and joint best among models with released weights, at $0.60 per Index task against GLM-5.3's $5.01. On the retired v4.3.2 scale it scored 45, a point behind Xiaomi's MiMo-V2.6-Pro.

GLM-5.3 ships under Z.ai's own glm-5.3 licence rather than MIT or Apache-2.0, and providers above $10bn in revenue need a Z.ai security review before commercial use, so read the LICENSE file in the repo before deploying. The weights landed on Hugging Face on 28 August, about two weeks after the August API launch, and the reign here is dated from 10 September. Artificial Analysis replaced v4.3 with Intelligence Index v5.0 on 27 October and re-based the whole table — GLM-5.3 (max) now reads 32 at $5.01 per Index task, its worst-in-class cost among comparable open-weight models — so any v4.x score or cost figure is no longer comparable.

days of reign

since 10 Sep 2026

Best small / on-device LLM

Gemma 4 26B A4B

A 25.2B-total MoE with only 3.8B active parameters, so it decodes at small-model speed, and Google's quantisation-aware training keeps 4-bit quality close to bfloat16. Be specific about which file you pull: Unsloth measured a naive Q4_0 conversion of the QAT checkpoint at 70.2% top-1 against 85.6% for their dynamic UD-Q4_K_XL build at 14.2GB, so prefer that over Google's own 14.4GB q4_0 GGUF, and budget the roughly 15GB of total memory it wants — it loads on a 16GB machine but leaves little room for KV cache, so plan for modest context rather than the full 256K. The pick is about deployability: Apache-2.0 weights, QAT builds sized for laptops, native function calling and image input, rather than raw benchmark score.

Challenger: MiniCPM5-2B — OpenBMB's official GGUF build is up, tagged for tool-calling and on-device use, with published file sizes — 1.56GB at Q4_K_M — so it clears a 16GB budget with room to spare. What is still missing is independent 4-bit retention data and any tool-calling-under-quantisation numbers. We have dropped the Artificial Analysis comparison for now: the index has re-based to v5.0 and the pick is unscored on it, so there is no like-for-like quality gap to quote; treat this as a deployability story rather than a quality one.

Qwen3.8 27B is still the quality leader in this size class — on Artificial Analysis's current Intelligence Index v5.0 it scores 20 at xhigh (17 medium, 16 low, 10 non-reasoning), while this pick carries no v5.0 score at all — but Qwen's only official quantised build is the FP8 repo, so a 16GB fit remains unproven, and the third-party options that do exist (community GGUF conversions from ggml-org and lmstudio-community, plus GPU-oriented NVFP4 checkpoints) come with no published retention data at a 16GB footprint. The pick's QAT int4 GGUF is 14.2GB in Unsloth's dynamic build, about 200MB smaller than a naive Q4_0 conversion of the same checkpoint — Google's own q4_0 file is 14.4GB — and wants roughly 15GB of total memory; that gap is the whole argument for it. The title stays under review because a verified 4-bit Qwen3.8 build with published file sizes and tool-calling numbers would likely take the slot.

213

days of reign

since 2 Apr 2026

Speech

2 titles

Best open-weight STT

ARK-ASR-3B

ARK-ASR-3B heads the Open ASR Leaderboard's public listing at 4.76 mean WER with an RTFx of 490.98, ahead of MOSS-Transcribe-preview-2B on 4.87, MOSS-Transcribe-Diarize on 5.17 and Cohere Transcribe on 5.42, with per-dataset results filed in the repo and dated 23 June 2026. The 5.04% on the card is the seven-set average that omits TEDLIUM; add TEDLIUM's 2.79% and you get the board figure. It ships Apache-2.0 with 19 languages, but that is not what separates it from MOSS: only the preview-2B is English-only, while MOSS-Transcribe-Diarize is also Apache-2.0 and covers 50+ languages, so ARK's edge there is accuracy rather than licence or reach.

Challenger: Voxtral Small Transcribe 3 — Launched on 20 October 2026 and still #3 on the Artificial Analysis non-streaming board at 2.1% AA-WER with a median speed factor of 62.2, behind Fun-Realtime-ASR-preview (1.7%) and MAI-Transcribe-2 (2.0%). That index averages only three datasets, so it will not line up with the Open ASR Leaderboard's eight-set mean, where this model does not appear and ARK-ASR-3B holds at 4.76. Artificial Analysis does not count it among its open-weights entries — its best open-weights row is the older Voxtral Small at 2.8% — so despite Mistral's claimed 50+ languages, the licence and weight availability this slot needs are still unverified.

NVIDIA's Parakeet-TDT-0.6B-v3 is still the sensible choice where throughput decides the invoice — 25 European languages, automatic language detection and a 6.34% average WER, so you are trading roughly 1.5 WER points for the speed — but note it ships under CC BY 4.0, not Apache-2.0. ARK's headline is board-verified rather than card-only: the leaderboard records 4.76 mean WER at RTFx 490.98, including TEDLIUM at 2.79%. Do budget for less throughput headroom than that suggests, though: AutoArk's own rerun of the seven public splits on 8x RTX 4090, scored with the leaderboard scorer, lands at 5.13% WER and an overall RTFx of 197.

132

days of reign

since 22 Jun 2026

Best open-weight TTS

BreezeBlue Breeze TTS 2 Open Weights

Top open-weight model on the Artificial Analysis provider-voice leaderboard at 1202 Elo, ninth overall and 82 points clear of the next open-weight model, Fish Audio S2 Pro at 1120. BreezeBlue quotes under 40ms time-to-first-audio and a 0.32 RTF on a warmed H100, but those numbers come from the --fast-all path, which wants about 14.4 GiB and a 24GB card; plain eager inference fits in roughly 7.7 GiB, so 12GB is the practical floor. English and Chinese only, and the weights sit under the BreezeBlue Research and Non-Commercial Licence, so commercial self-hosting still needs a separate arrangement.

Challenger: StepFun StepAudio 2.5 TTS (Aug 2026) — Now trails the pick on the Artificial Analysis provider-voice board, 1214 to 1228 — 18th against 13th — on the board's current scale. It stays proprietary and API-only on StepFun's platform at about $85 per 1M characters with no weights published, so it is not an option for self-hosters either way. StepFun's open-weight audio line remains Step-Audio-EditX and Step-Audio-2-mini.

Open-weight but not open-source: the inference code is Apache-2.0, while the weights, derivatives and self-hosted outputs fall under the BreezeBlue Research and Non-Commercial Licence, and a paid breezeblue.ai subscription only covers outputs from the hosted API, not anything you generate locally. If you need commercial self-hosting, MIT-licensed Ming-3.0-tts (1187 on the provider-voice board) and Fish Audio Drama 3 (Community Licence v1.0, free commercially under US$500k annual revenue, 1278) are now better fallbacks than Apache-2.0 Step Audio EditX at 1099. The standing is board-specific and language-limited too: on the controlled-voice arena Breeze scores 1021, fourth among open-weight models behind Drama 3 (1085), Ming-3.0-tts (1068) and Drama 3 Preview (1063), and the model speaks English and Chinese only.

days of reign

since 9 Sep 2026

Create

2 titles

Best image generation

New titleholder

GPT-Image-2.5-Sunburst

Top of Artificial Analysis's text-to-image board at 1,222, fifteen clear of stablemate GPT Image 2.5 Flare on 1,207 and 36 clear of GPT Image 2 (high) on 1,186, and top of the same site's editing board at 1,199 — though MAI-Image-2.7-Preview is only six points back on 1,193 and statistically tied with it. It also leads both Arena boards, text-to-image 1,425 against Flare's 1,398 and single-image edit 1,517 against Flare's 1,482, with both entries still flagged preliminary despite tens of thousands of votes. At $210.70 per 1,000 images it is no dearer than the model it replaces, but MAI-Image-2.6 ($38.90) remains far cheaper for editing at volume.

Editing is no longer a GPT Image 2 story: on Arena's single-image edit board it sits fourth on 1,460, behind GPT-Image-2.5-Sunburst (1,517), GPT-Image-2.5-Flare (1,482) and MAI-Image-2.7-Preview (1,471), with Grok Imagine Image 2.0 fifth on 1,439 and MAI-Image-2.6 sixth on 1,427. Artificial Analysis's editing board now has it eighth on 1,136, behind Sunburst (1,199), MAI-Image-2.7-Preview (1,193), Flare (1,182), a three-way tie between MAI-Image-2.6, Krea 3 (Ultra) and Nano Banana 2.1 on 1,151, and MAI-Image-2.6-Flash (1,144). On price it is level with the newer OpenAI models rather than uniquely dear, at $211 per 1,000 images against $210.70 for both GPT-Image-2.5 variants, though still far above $67 for Nano Banana 2 and $38.90 for MAI-Image-2.6.

day of reign

since 31 Oct 2026

Best video generation

New titleholder

Alibaba Wan 3.0

Wan 3.0 is #1 on Artificial Analysis's text-to-video with-audio board (1175, with the previous pick, Gemini Omni Flash 1.1, tenth on 1128) and #1 on the video editing board (1196), though on the silent text-to-video board it is only fourth (1133) behind HeyGen Video 1 (1143). Alibaba's Model Studio docs confirm 2-30 second clips in a single job at up to 1080p and 30fps, with dialogue, sound effects and music generated natively in the same pass. It is hosted-only with no open weights, so confirm Alibaba Cloud's terms before commercial work.

The lead is narrower than it looks: Wan 3.0 tops text-to-video with audio (1175) and video editing (1196) on Artificial Analysis, but sits fourth on the silent text-to-video board (1133, behind HeyGen Video 1 on 1143) and ninth on image-to-video with audio (1111), where Dreamina Seedance 2.5 leads on 1153. Note the boards were re-based for v2.1, with the scale pinned so Kling 3.0 1080p (Pro) sits at 1000, so Elo figures quoted earlier this year are not comparable. There are no open weights either: Wan 3.0 is hosted on Alibaba Cloud Model Studio and listed at $12 per minute, so the Wan-AI Hugging Face org carries only the older Wan 2.x releases.

day of reign

since 31 Oct 2026

Build

2 titles

Best text embeddings

Nemotron-3-Embed-8B

NVIDIA's launch materials put Nemotron-3-Embed-8B-BF16 top of the RTEB multilingual board as of 16 July 2026, on a headline 78.46 average NDCG@10 across the 16 public RTEB tasks its model card actually reports, plus 75.45 on MMTEB Retrieval. The card confirms 34 evaluated languages, a 32,768-token maximum sequence length and 4,096-dimension vectors you can slice to 2,048 or 1,024 and re-normalise, with weights under OpenMDW-1.1 and 1B BF16 and NVFP4 checkpoints for cheaper serving. Mind the basis before you quote it: the live RTEB(beta) multilingual board lists 30 tasks over 22 languages, ranks by Borda count of per-task ranks rather than by mean score, and has had its private column temporarily removed since January 2026, so the placing rests on a vendor snapshot of an open-task view.

Challenger: EmbeddingGemma 2 — Google's open, lightweight EmbeddingGemma 2 was announced on 6 October 2026: a 740M form factor with modular encoders that maps text, code, images, video and audio into a single embedding space, aimed squarely at on-device work. It is no longer wholly untested against our pick, as NVIDIA's 21 October Nemotron 3.5 post benchmarks EmbeddingGemma-2 head-to-head on ViDoRe v3, Korean ViDoRe and MIRACL Vision, but those are visual retrieval runs. There is still no like-for-like RTEB or MTEB retrieval comparison against Nemotron-3-Embed-8B, so for pure text retrieval the gap remains unmeasured.

Microsoft's card for harrier-oss-v1-27b claims state-of-the-art on Multilingual MTEB v2 at release, with a 74.3 average across a 94-language family (0.6b 69.0, 270m 66.5), but it gives no task count, no rival scores and no Borda ranking — so treat any head-to-head against Tencent's KaLM-Embedding-Gemma3-12B-2511 as unconfirmed rather than settled. That card dates from late March 2026 and the model landed in Foundry in April, so pull the live MTEB table before repeating either number. Either way an MTEB v2 task mean and our pick's RTEB NDCG@10 answer different questions: one aggregates all task types, the other is retrieval only.

108

days of reign

since 16 Jul 2026

Best inference serving

vLLM

Spheron's like-for-like H100 runs on Llama 3.3 70B FP8 (three-engine post published March 2026, two-engine June 2026, both refreshed 19 October 2026) put SGLang roughly 1-5% ahead of vLLM on unique-prompt traffic, a gap Spheron itself calls run-to-run variance. Both stacks are now well out of date: they used vLLM v0.18.0 on the legacy model runner, before Model Runner V2 became the default for all dense models in v0.25.0 (11 July 2026) and for every model in v0.29.0 (9 September 2026), and SGLang v0.5.9, before breakable CUDA graphs became SGLang's default prefill path in v0.5.15. Pick vLLM for breadth, hardware coverage and release discipline rather than for any current throughput number.

Challenger: SGLang — Near-parity overall and faster on agentic/structured workloads, with major production adoption (xAI, AMD, LinkedIn, Cursor).

Less of a two-horse race than it looks: in Spheron's three-engine H100 run on Llama 3.3 70B FP8, TensorRT-LLM led vLLM by 8-13% at 1-50 concurrent requests and about 16% at 100 once its engine was compiled, but it takes roughly 28 minutes to first request against about a minute for vLLM and SGLang. SGLang's edge is prefix-driven rather than general: with an 80% shared 512-token prefix it cut TTFT p50 by 37% at 50 concurrent requests, while on unique prompts the two sat within a few per cent, and enabling vLLM's --enable-prefix-caching narrowed the TTFT gap to roughly 15-18% at 10 concurrent requests, though at 50 Spheron still puts APC-on vLLM at 2,100 effective tok/s against SGLang's 2,550. Measure your own prefix overlap and re-run on current releases: those figures predate both vLLM's Model Runner V2 default (all dense models from v0.25.0, every model from v0.29.0, with v0.32.0 current) and SGLang's breakable CUDA graph default from v0.5.15.

1230

days of reign

since 20 Jun 2023