# Search and deep research leaderboard

As of November 21, 2026, Claude Fable 5.2 leads our search and deep research leaderboard at 86.6 (Supported, $8/$40 per 1M tokens), ahead of GPT-6.2 Astra at 88.6. Gemini 4 Carbon (#3, 83.7) is the lowest-cost model near the top at $2/$10 per 1M tokens; MiMo-V2.6-Pro leads open-weight models at #15 (69.0). We rank 74 models (36 Supported, 38 Estimated) across 9 domain protocols, and this vertical carries zero weight in the overall score.

## Shortlist by workload and budget

- **Highest point score:** [Claude Fable 5.2](/models/claude-fable-5-2) (#1, 86.6, Supported, $8/$40 per 1M tokens) — Costs $8/$40 per 1M tokens; Gemini 4 Carbon (#3, 83.7) cuts token spend to $2/$10 per 1M tokens.
- **Closest contender:** [GPT-6.2 Astra](/models/gpt-6-2-astra) (#2, 88.6, Estimated, $8/$40 per 1M tokens) — Sits -2.0 points behind Claude Fable 5.2; test both on your own prompts when provider lock-in or latency matters.
- **Best API value:** [Gemini 4 Carbon](/models/gemini-4-carbon) (#3, 83.7, Supported, $2/$10 per 1M tokens) — Keeps 97% of the top score (83.7 vs 86.6), but gives up 2.9 points on the hardest tail cases.
- **Top open-weight model:** [MiMo-V2.6-Pro](/models/mimo-v2-6-pro) (#15, 69.0, Estimated, Open weights · $0.43/$0.87 per 1M tokens) — Trails Claude Fable 5.2 by 17.6 points (#15 overall here), so pick it when weight control or on-prem residency outweighs peak score.

## Ranked models

| Rank | Model | Creator | Score | Conditional range | Evidence | API ($/1M in/out) | Context | Source families |
|---:|---|---|---:|---:|---|---:|---:|---:|
| #1 | [Claude Fable 5.2](/models/claude-fable-5-2) | Anthropic | 86.6 | 75.7-97.6 | Supported | $8.00 / $40.00 | 1M | 4 |
| #2 | [GPT-6.2 Astra](/models/gpt-6-2-astra) | OpenAI | 88.6 | 76.7-100.0 | Estimated | $8.00 / $40.00 | 1.05M | 1 |
| #3 | [Gemini 4 Carbon](/models/gemini-4-carbon) | Google | 83.7 | 73.3-94.2 | Supported | $2.00 / $10.00 | 1M | 4 |
| #4 | [GPT-6 Astra](/models/gpt-6-astra) | OpenAI | 78.2 | 68.0-88.4 | Supported | $10.00 / $50.00 | 1.05M | 5 |
| #5 | [Claude Opus 5.5](/models/claude-opus-5-5) | Anthropic | 76.7 | 66.6-86.9 | Supported | $4.00 / $20.00 | 1M | 5 |
| #6 | [GPT-6.1 Sol](/models/gpt-6-1-sol) | OpenAI | 76.6 | 66.2-87.0 | Supported | $2.00 / $10.00 | 1.05M | 4 |
| #7 | [GPT-6.1 Terra](/models/gpt-6-1-terra) | OpenAI | 76.4 | 66.0-86.8 | Supported | $1.00 / $5.00 | 1.05M | 4 |
| #8 | [Gemini 4 Argon](/models/gemini-4-argon) | Google | 75.3 | 65.1-85.4 | Supported | $2.00 / $10.00 | 1M | 4 |
| #9 | [Gemini 4 Hydrogen](/models/gemini-4-hydrogen) | Google | 75.0 | 64.5-85.5 | Supported | $1.50 / $7.50 | 1M | 3 |
| #10 | [Claude Fable 5.1](/models/claude-fable-5-1) | Anthropic | 73.5 | 63.0-84.0 | Supported | $10.00 / $50.00 | 1M | 4 |
| #11 | [Claude Sonnet 5.5](/models/claude-sonnet-5-5) | Anthropic | 73.3 | 60.8-85.7 | Supported | $2.00 / $10.00 | 1M | 4 |
| #12 | [Qwen4 Max](/models/qwen4-max) | Alibaba | 72.0 | 58.0-86.0 | Estimated | $1.50 / $4.50 | 1M | 2 |
| #13 | [GPT-6 Sol](/models/gpt-6-sol) | OpenAI | 71.4 | 60.6-82.3 | Supported | $2.00 / $10.00 | 1.05M | 2 |
| #14 | [GPT-5.6 Sol](/models/gpt-5-6-sol) | OpenAI | 69.2 | 59.0-79.5 | Supported | $4.00 / $20.00 | 1.05M | 3 |
| #15 | [MiMo-V2.6-Pro](/models/mimo-v2-6-pro) | Xiaomi | 69.0 | 54.1-84.0 | Estimated | $0.43 / $0.87 | 1M | 2 |
| #16 | [DeepSeek V4.1 Flash 1017](/models/deepseek-v4-1-flash) | DeepSeek | 68.3 | 55.9-80.7 | Supported | $0.30 / $1.20 | 1M | 3 |
| #17 | [Claude Opus 5](/models/claude-opus-5) | Anthropic | 68.3 | 58.1-78.4 | Supported | $5.00 / $25.00 | - | 5 |
| #18 | [MiniMax M3.1 Flash](/models/minimax-m3-1-flash) | MiniMax | 68.2 | 52.9-83.5 | Estimated | $0.15 / $0.60 | 1M | 2 |
| #19 | [Kimi K3.1 Pro](/models/kimi-3-1-pro) | Moonshot AI | 68.1 | 57.2-79.0 | Supported | $2.00 / $10.00 | 1.05M | 4 |
| #20 | [Kimi K3](/models/kimi-k3) | Moonshot AI | 67.8 | 56.9-78.8 | Supported | $3.00 / $15.00 | 1.05M | 4 |
| #21 | [Claude Haiku 5.5](/models/claude-haiku-5-5) | Anthropic | 67.4 | 57.1-77.6 | Supported | $0.10 / $0.50 | 1M | 4 |
| #22 | [Claude Fable 5](/models/claude-fable) | Anthropic | 66.0 | 55.7-76.4 | Estimated | $10.00 / $50.00 | 1M+ | 2 |
| #23 | [GPT-5.6 Terra](/models/gpt-5-6-terra) | OpenAI | 65.4 | 54.7-76.1 | Supported | $2.00 / $12.00 | 1.05M | 4 |
| #24 | [Grok 4.6](/models/grok-4-6) | xAI | 64.3 | 54.1-74.6 | Supported | $2.00 / $6.00 | 500K | 4 |
| #25 | [GPT-6 Luna](/models/gpt-6-luna) | OpenAI | 64.0 | 51.0-77.1 | Supported | $0.10 / $0.50 | 1.05M | 3 |
| #26 | [GPT-5.5](/models/gpt-5-5) | OpenAI | 63.2 | 52.6-73.8 | Supported | $5.00 / $30.00 | 1M | 3 |
| #27 | [Step 5 Preview](/models/step-5-preview) | StepFun | 62.7 | 48.5-76.9 | Estimated | $1.00 / $2.70 | 1M | 4 |
| #28 | [Qwen3.8 Max](/models/qwen3-8-max) | Alibaba | 62.6 | 50.2-74.9 | Supported | - | 1M | 3 |
| #29 | [GLM-5.3](/models/glm-5-3) | Z.AI | 64.5 | 49.4-79.5 | Estimated | Free | 1M | 1 |
| #30 | [GLM-5.3-Flash](/models/glm-5-3-flash) | Z.AI | 62.3 | 49.6-75.1 | Estimated | Free | 1M | 2 |
| #31 | [Claude Opus 4.8](/models/claude-opus-4-8) | Anthropic | 62.3 | 52.3-72.2 | Supported | $5.00 / $25.00 | 1M | 4 |
| #32 | [Muse Wonder](/models/muse-wonder) | Meta | 63.9 | 52.6-75.2 | Estimated | $4.00 / $20.00 | 1M | 1 |
| #33 | [Gemini 3.8 Flash](/models/gemini-3-8-flash) | Google | 61.7 | 49.2-74.3 | Supported | $0.75 / $3.75 | 1M | 3 |
| #34 | [GPT-5.6 Luna](/models/gpt-5-6-luna) | OpenAI | 61.2 | 50.0-72.4 | Estimated | $0.20 / $1.20 | 1.05M | 2 |
| #35 | [GPT-5.4](/models/gpt-5-4) | OpenAI | 61.0 | 50.6-71.4 | Estimated | $2.50 / $15.00 | 1.05M | 3 |
| #36 | [Claude Opus 4.7](/models/claude-opus-4-7) | Anthropic | 58.9 | 48.3-69.5 | Supported | $5.00 / $25.00 | 1M | 3 |
| #37 | [Claude Opus 4.6](/models/claude-opus-4-6) | Anthropic | 58.7 | 48.0-69.3 | Supported | $5.00 / $25.00 | 1M | 4 |
| #38 | [Claude Opus 4.6 (Adaptive)](/models/claude-opus-4-6-thinking) | Anthropic | 58.7 | 48.0-69.3 | Supported | - | 1M | 4 |
| #39 | [Gemini 3.1 Pro](/models/gemini-3-1-pro) | Google | 58.5 | 48.0-69.1 | Supported | $2.00 / $12.00 | 1M | 3 |
| #40 | [Gemini 3.7 Flash](/models/gemini-3-7-flash) | Google | 60.1 | 45.1-75.1 | Estimated | $0.75 / $3.75 | 1M | 1 |
| #41 | [Muse Spark Open](/models/muse-spark-open) | Meta | 55.9 | 40.2-71.6 | Estimated | - | 1M | 2 |
| #42 | [Claude Opus 4.5](/models/claude-opus-4-5) | Anthropic | 55.8 | 44.0-67.5 | Supported | $5.00 / $25.00 | 200K | 3 |
| #43 | [Claude Opus 4.5 Thinking](/models/claude-opus-4-5-thinking) | Anthropic | 55.8 | 44.0-67.5 | Supported | - | 200K | 3 |
| #44 | [GPT-5](/models/gpt-5) | OpenAI | 55.7 | 42.9-68.6 | Supported | $1.25 / $10.00 | 128K | 2 |
| #45 | [Kimi K2.6](/models/kimi-2-6) | Moonshot AI | 55.2 | 44.6-65.9 | Estimated | $0.95 / $4.00 | 256K | 4 |
| #46 | [Grok 4.5](/models/grok-4-5) | xAI | 56.8 | 41.8-71.7 | Estimated | $2.00 / $6.00 | 500K | 1 |
| #47 | [Inkling](/models/inkling) | Thinking Machines Lab | 53.6 | 42.5-64.7 | Estimated | $1.87 / $4.68 | 1M | 3 |
| #48 | [Qwen3.7 Max](/models/qwen3-7-max) | Alibaba | 55.5 | 44.8-66.3 | Estimated | - | 1M | 1 |
| #49 | [GLM-5.2](/models/glm-5-2) | Z.AI | 55.5 | 44.6-66.3 | Estimated | $1.40 / $4.40 | 1M | 1 |
| #50 | [Qwen3.8-27B](/models/qwen3-8-27b) | Alibaba | 52.9 | 39.8-66.0 | Estimated | Free | 262K | 2 |
| #51 | [Gemini 3.5 Flash](/models/gemini-3-5-flash) | Google | 52.4 | 41.9-62.9 | Supported | $1.50 / $9.00 | 1M | 2 |
| #52 | [Gemini 3.6 Flash](/models/gemini-3-6-flash) | Google | 54.3 | 39.2-69.3 | Estimated | $0.75 / $3.75 | 1M | 1 |
| #53 | [GPT-5.2](/models/gpt-5-2) | OpenAI | 51.4 | 40.1-62.7 | Supported | $1.75 / $14.00 | 400K | 3 |
| #54 | [o3](/models/o3) | OpenAI | 52.8 | 35.0-70.6 | Estimated | $2.00 / $8.00 | 200K | 1 |
| #55 | [Claude Sonnet 4.6](/models/claude-sonnet-4-6) | Anthropic | 52.3 | 35.2-69.5 | Estimated | $3.00 / $15.00 | 200K | 1 |
| #56 | [Claude Sonnet 4.5](/models/claude-sonnet-4-5) | Anthropic | 49.9 | 37.4-62.5 | Supported | $3.00 / $15.00 | 200K | 2 |
| #57 | [Claude Sonnet 4.5 Thinking](/models/claude-sonnet-4-5-thinking) | Anthropic | 49.9 | 37.4-62.5 | Supported | - | 200K | 2 |
| #58 | [Grok 4](/models/grok-4) | xAI | 50.5 | 32.5-68.5 | Estimated | - | 128K | 1 |
| #59 | [Grok 4.1 Fast (Reasoning)](/models/grok-4-1-fast-reasoning) | xAI | 48.7 | 27.1-70.3 | Estimated | - | 2M | 1 |
| #60 | [GPT-5.1](/models/gpt-5-1) | OpenAI | 46.4 | 28.3-64.6 | Estimated | $1.25 / $10.00 | 200K | 1 |
| #61 | [Grok 4 Fast (Reasoning)](/models/grok-4-fast-reasoning) | xAI | 45.1 | 25.7-64.4 | Estimated | - | 2M | 1 |
| #62 | [Gemma 4 31B](/models/gemma-4-31b) | Google | 43.0 | 30.0-56.0 | Supported | Free | 256K | 3 |
| #63 | [Mistral Large 4](/models/mistral-large-4) | Mistral | 42.7 | 28.3-57.1 | Estimated | $0.70 / $2.10 | 1M | 2 |
| #64 | [Gemini 3.5 Flash-Lite](/models/gemini-3-5-flash-lite) | Google | 43.5 | 28.4-58.5 | Estimated | $0.30 / $2.50 | 1M | 1 |
| #65 | [GPT-5 mini](/models/gpt-5-mini) | OpenAI | 43.4 | 23.7-63.1 | Estimated | $0.25 / $2.00 | 128K | 1 |
| #66 | [Gemini 3.1 Flash-Lite](/models/gemini-3-1-flash-lite) | Google | 43.2 | 24.3-62.1 | Estimated | $0.25 / $1.50 | 1M | 1 |
| #67 | [Gemini 2.5 Pro](/models/gemini-2-5-pro) | Google | 42.7 | 24.6-60.7 | Estimated | $1.25 / $10.00 | 1M | 1 |
| #68 | [Gemini 2.5 Flash](/models/gemini-2-5-flash) | Google | 41.9 | 23.0-60.8 | Estimated | $0.30 / $2.50 | 1M | 1 |
| #69 | [GPT-4.1](/models/gpt-4-1) | OpenAI | 39.7 | 21.9-57.5 | Estimated | $2.00 / $8.00 | 1M | 1 |
| #70 | [Gemma 4 26B A4B](/models/gemma-4-26b-a4b) | Google | 37.2 | 22.7-51.7 | Estimated | Free | 256K | 1 |
| #71 | [GPT-4o](/models/gpt-4o) | OpenAI | 33.4 | 14.0-52.7 | Estimated | $2.50 / $10.00 | 128K | 1 |
| #72 | [Gemma 3 27B](/models/gemma-3-27b) | Google | 29.4 | 8.5-50.2 | Estimated | Free | 32K | 1 |
| #73 | [GPT-OSS 120B](/models/gpt-oss-120b) | OpenAI | 27.3 | 10.4-44.1 | Estimated | Free | 128K | 1 |
| #74 | [GPT-OSS 20B](/models/gpt-oss-20b) | OpenAI | 27.0 | 5.3-48.7 | Estimated | Free | 128K | 1 |

## Scored domain benchmarks (9)

- [FACTS Search V2](/benchmarks/facts-search) (17% ref. weight): FACTS Search V2 measures factual answering with a shared Brave Search tool. Overall, public, private, hedging, search-count, attempted-accuracy, and F1 measurements remain distinct.
- [EBR-bench](/benchmarks/ebrbench) (16% ref. weight): Epoch AI's EBR-bench tests whether models get better at an unfamiliar game with practice: they play the campaign card game Earthborne Rangers repeatedly, take notes, and are scored on later playthroughs.
- [BrowseComp](/benchmarks/browsecomp) (13% ref. weight): BrowseComp tests whether a browsing agent can find difficult, fact-seeking answers across the web. Its 1,266 problems use short answers that are easier to grade than an open-ended research report.
- [WideResearch](/benchmarks/wideresearch) (11% ref. weight): A broad research-agent benchmark for open-ended information gathering, synthesis, and answer construction across wide search spaces.
- [LiveBench 2.0 Research](/benchmarks/livebench) · LiveBench (10% ref. weight): Multi-step information synthesis and source verification on fresh LiveBench tasks.
- [LiveBench 2.0 Browser](/benchmarks/livebench) · LiveBench (10% ref. weight): Interactive browser navigation and page-state extraction on fresh LiveBench web tasks.
- [DeepSearchQA](/benchmarks/deepsearchqa) (8% ref. weight): An agentic browsing benchmark where models search the web, gather evidence, and answer list-style questions using browser tools.
- [DRACO](/benchmarks/draco) (7% ref. weight): Agentic data-analysis tasks scored against per-task rubrics at a 980K-token budget.
- [FACTS Search V1](/benchmarks/facts-search) (7% ref. weight): The archived FACTS Search V1 protocol measures factual answering with Brave Search and a Gemini 2.0 Flash grader. Its scores remain separate from Search V2.

## Multi-hop web browsing, enterprise retrieval, and citation-backed synthesis

Nine search and research benchmarks power this board, led by FACTS Search V2 (17%), EBR-bench (16%), BrowseComp (13%) alongside WideResearch, LiveBench 2.0 Research and Browser, DeepSearchQA, and DRACO. These evaluations test whether an agent can decompose ambiguous multi-constraint queries, reformulate failed searches, reconcile conflicting web pages, and synthesize long-form reports whose citations actually support each sentence.

Thirty-four of the 74 ranked models carry Supported status, giving buyers multi-evaluator evidence across Google DeepMind's FACTS Search suite, BrowseComp, EBR-bench, and LiveBench.

## Your search index, page parser, and retry cap move accuracy as much as the LLM

A search score never belongs to model weights in isolation. Switching from SerpAPI to an internal Elasticsearch cluster, stripping JavaScript tables during HTML-to-Markdown conversion, or capping an agent at five tool calls instead of twenty will reorder close models immediately.

Benchmark snapshots also cannot guarantee live breaking-news freshness or paywall access in your own environment; evaluate candidate models inside your exact retrieval and reranking stack before shipping customer-facing answers.

[Web-research model guide](/best/web-research) · [Claude Fable 5.2 vs GPT-6.2 Astra](/compare/claude-fable-5-2-vs-gpt-6-2-astra)

## Choosing between deep-research flagships and fast retrieval workhorses

Claude Fable 5.2 (86.6, Estimated, $8/$40 per 1M tokens) leads GPT-6.2 Astra (88.6, Supported, $8/$40 per 1M tokens) by -2.0 points, making GPT-6.2 Astra the highest Supported choice when you want confirmed multi-evaluator evidence. Because deep-research loops often read tens of thousands of retrieved tokens per query, input cost compounds fast.

At volume, Gemini 4 Carbon (#3, 83.7, Supported at $2/$10 per 1M tokens) cuts token spend by 75% while giving up 2.9 points, and MiMo-V2.6-Pro (#15, 69.0) leads self-hosted weights for private intranet search.

[Agentic leaderboard](/agentic) · [Enterprise search AI comparison](/best/enterprise-search-ai)

## Questions

### Which LLM is best for web search and deep research right now?

Claude Fable 5.2 holds #1 on our web search and deep research leaderboard at 86.6 (Supported) as of November 21, 2026, ahead of GPT-6.2 Astra at 88.6 and Gemini 4 Carbon at 83.7. Both top models hold Supported status across independent public evaluations.

### What is the best open-weight LLM for web search and deep research?

MiMo-V2.6-Pro from Xiaomi leads open-weight models on this web search and deep research board at #15 with a calibrated score of 69.0 (Estimated). That trails Claude Fable 5.2 by 17.6 points, so choose MiMo-V2.6-Pro when self-hosting, VPC data residency, or weight customization matters more than peak frontier accuracy.

### Which API model offers the best value for web search and deep research?

Gemini 4 Carbon (#3, 83.7) is our top API value pick for web search and deep research, priced at $2/$10 per 1M tokens compared with Claude Fable 5.2 at $8/$40 per 1M tokens. You give up 2.9 calibrated points, making Gemini 4 Carbon the better buy whenever those extra points do not cut downstream retries.

### Which benchmarks make up the web search and deep research leaderboard score?

BenchAlign v6.0 calibrates 9 open and primary domain protocols for this board, led by FACTS Search V2 (17%), EBR-bench (16%), BrowseComp (13%), WideResearch (11%). Each percentage reflects share of the vertical's reference weight inside our item-response model rather than a raw score average, and no single independent evaluator can exceed 40% when three or more evaluators contribute.

### How much does the browsing harness affect search leaderboard scores?

Substantially. FACTS Search, BrowseComp, EBR-bench, and WideResearch each pair models with a specific search index, page-fetch parser, and step limit. Changing search APIs, stripping HTML tables, or lowering tool-call budgets can flip two-point leads, so benchmark your top two finalists inside your own retrieval pipeline.

### How does the search leaderboard differ from the factuality leaderboard?

Our factuality board tests closed-book parametric accuracy and faithfulness to static passages already in the prompt. This search board tests active query decomposition, multi-hop web navigation, enterprise index retrieval, and cross-document synthesis across nine agentic research protocols, carrying zero direct weight in the overall score.

## Related benchmarks and guides

- [Agentic Leaderboard](/agentic): Broader tool use, terminal work, and multi-step execution rankings.
- [Best LLMs for Web Research](/best/web-research): Shortlist for browsing agents, source gathering, and cited briefs.
- [Enterprise Search AI](/best/enterprise-search-ai): Compare retrieval architectures and grounded enterprise search stacks.
- [Factuality Leaderboard](/leaderboards/factuality): Source-grounded verification and abstention rankings without live web access.

## Other specialist vertical leaderboards

- [Cybersecurity leaderboard](/leaderboards/cybersecurity) — #1 Gemini 4 Carbon (90.0)
- [Documents and extraction leaderboard](/leaderboards/documents) — #1 GPT-6.2 Astra (86.5)
- [Factuality and anti-hallucination leaderboard](/leaderboards/factuality) — #1 Gemini 4 Carbon (78.8)
- [Finance and accounting leaderboard](/leaderboards/finance) — #1 Gemini 4 Carbon (89.2)
- [Frontend and web development leaderboard](/leaderboards/frontend) — #1 Claude Fable 5.2 (89.5)
- [Healthcare and medical leaderboard](/leaderboards/healthcare) — #1 Claude Fable 5.2 (87.5)
- [Legal leaderboard](/leaderboards/legal) — #1 Claude Fable 5.2 (86.1)
- [Science and research leaderboard](/leaderboards/science) — #1 Claude Fable 5.2 (94.1)

Last updated: November 21, 2026

Canonical page: https://benchlm.ai/leaderboards/search
