# Hello, AI
> Curated directory of frontier AI models with LMArena text Elo, list prices and context windows, plus task-specific recommendations. Updated weekly (2026-10-25).

## API (no auth, 100 requests per minute per IP)
- [OpenAPI spec](https://helloai.com/api/openapi.json)
- [Status and data freshness](https://helloai.com/api/status)
- [All models](https://helloai.com/api/models)
- [Recommend](https://helloai.com/api/recommend?task=coding): params task, max_cost, min_context, provider, limit. `score` is an ordering aid, not a quality measure: compare scores only between calls that share the same `meta.scoring` version, snapshot and resolved task.
- [MCP server](https://helloai.com/mcp): Streamable HTTP, no auth. Tools recommend_models, list_models, get_model, list_categories, list_articles, get_status. Claude Code: `claude mcp add --transport http helloai https://helloai.com/mcp`.

## Notes
- Elo is the LMArena (arena.ai) text-overall board, recorded per model in `elo_source`. A model described as not yet rated, missing or stale has no usable Elo of its own and is not ranked.

## Disclosures
Hello, AI is run by Clement Machado. I have no financial or employment relationship with Anthropic, Google, xAI, Meta, Alibaba or OpenAI: no equity, no payments from them, no sponsorship, and no free credits or early access. The weekly update and the article drafts are produced with the help of AI models, including Claude and Grok, so two of the vendors on the list are also tools I use. That is a reason to check my work against the published method. I also run the HelloAI app at app.helloai.com, a separate product. Links to it from this site are labelled "our app". The model rankings are not affected by it.

## Models tracked
- Claude Opus 5.5 (Anthropic): Elo 1508 ± 7 (high); $4/M in, $20/M out; 1M ctx; https://claude.ai
- Claude Fable 5.1 (Anthropic): Elo 1502 ± 6 (max); $10/M in, $50/M out; 1M ctx; https://claude.ai
- Muse Spark 1.3 (Meta): Elo 1494 ± 6 (max); $1.25/M in, $4.25/M out; 1M ctx; https://www.meta.ai/
- Gemini 3.1 Pro (Google): Elo 1487 ± 3; $2/M in, $12/M out; 1M ctx; https://gemini.google.com
- Qwen3.8-Max (Alibaba): Elo 1482 ± 5; $2/M in, $6/M out; 1M ctx; https://www.qwencloud.com/models/qwen3.8-max
- Grok 4.7 (xAI): Elo 1444 ± 6 (xhigh); $2/M in, $6/M out; 500K ctx; https://grok.com

## Articles
- Haiku 5.5 Costs $0.10 In. Muse Spark's Budget Crown Is Gone: https://helloai.com/articles/claude-haiku-5-5-ten-cents-input (2026-10-25)
- OpenAI Agents Escaped an Eval Sandbox. The Hole Was a Shared Cache: https://helloai.com/articles/openai-agents-escaped-eval-sandbox-shared-cache (2026-10-17)
- Two $2/$10 Launches, One Clear Winner on the Index: https://helloai.com/articles/two-dollar-tier-sonnet-5-5-vs-gpt-6-1-sol (2026-10-12)
- Three Agents, One Repo: Peer Tests Beat Fixed Suites: https://helloai.com/articles/head-to-head-agent-cup-peer-tests-beat-fixed-suites (2026-10-04)
- Opus 5.5's Own Text Score Takes the Lead: https://helloai.com/articles/opus-5-5-own-score-takes-the-lead (2026-09-30)
- Opus 5.5 Cuts the Price. Grok 4.7 Keeps Its Rate: https://helloai.com/articles/opus-5-5-cuts-price-grok-4-7-same-rate (2026-09-23)
- Astra Hits the Text Board. The Two-Week Clock Is Still Ticking: https://helloai.com/articles/gpt-6-astra-text-elo-lands-two-week-clock (2026-09-18)
- Astra Takes WebDev. helloai's Elo Gate Still Holds: https://helloai.com/articles/gpt-6-astra-takes-webdev-elo-gate-holds (2026-09-07)
- Fable 5.1 Cuts Cache Cost While Taking WebDev: https://helloai.com/articles/claude-fable-5-1-cheaper-cache-webdev-lead (2026-09-03)
- OpenAI Cuts Sol's Price 20% While It Leads Terminal-Bench: https://helloai.com/articles/openai-gpt-5-6-sol-price-cut-terminal-bench-lead (2026-08-30)

## Archived articles (prices and rankings as of each date)
- Grok Bot Is a Cloud Teammate. The Local CLI Runs the Site: https://helloai.com/articles/grok-bot-cloud-teammate-local-cli (2026-08-26)
- Qwen3.8-Max Cools Below Gemini as Arena Votes Arrive: https://helloai.com/articles/qwen3-8-max-cools-below-gemini (2026-08-22)
- DeepSeek Doubled Its Rate Card and Still Leads Daily Use: https://helloai.com/articles/deepseek-v4-pro-price-hike-still-cheapest (2026-08-16)
- Grok 4.6 Ships the Week After helloai Dropped It: https://helloai.com/articles/grok-4-6-ships-after-helloai-drop (2026-08-12)
- Qwen3.8-Max and DeepSeek Join; Grok and Kimi Exit: https://helloai.com/articles/qwen-deepseek-gptoss-rebalance-frontier-set (2026-08-07)
- Muse Spark 1.2 Ships With Muse Code: https://helloai.com/articles/muse-spark-1-2-muse-code-agentic-upgrade (2026-08-07)
- Claude Opus 5 Just Beat Anthropic's Own Flagship: https://helloai.com/articles/claude-opus-5-outscores-fable-5-at-half-price (2026-07-25)
- Gemini 3.6 Flash Ships — 3.5 Pro Still Waiting: https://helloai.com/articles/gemini-3-6-flash-ships-3-5-pro-still-waiting (2026-07-22)
- Kimi K3 Replaces GPT-5.6 Sol in helloai's Frontier Set: https://helloai.com/articles/kimi-k3-replaces-gpt-in-helloai-set (2026-07-17)
- ChatGPT Work vs Claude Cowork: The Agentic Seat War: https://helloai.com/articles/chatgpt-work-vs-claude-cowork-agentic-seat-war (2026-07-17)
- Muse Spark 1.1 Replaces Qwen3.7-Max in helloai's Frontier Set: https://helloai.com/articles/muse-spark-1-1-replaces-qwen-in-helloai-set (2026-07-09)
- Grok 4.5 Ships as xAI's Default Chat Model: https://helloai.com/articles/grok-4-5-ships-as-xai-default (2026-07-09)
- GPT-5.6 Sol Reaches General Availability: https://helloai.com/articles/gpt-5-6-sol-general-availability (2026-07-09)
- Ollama, Hermes, and GLM-5.2:cloud Are Rewriting the Margin Map: https://helloai.com/articles/ollama-hermes-glm-5-2-cloud-frontier-margins (2026-07-04)
- How Caching and Batching Cut Frontier API Costs by 90%: https://helloai.com/articles/how-caching-and-batching-cut-frontier-costs-90-percent (2026-07-04)
- Frontier Models Score Under 1% on ARC-AGI-3: https://helloai.com/articles/arc-agi-3-frontier-models-under-1-percent (2026-07-04)
- Frontier Releases Now Run Through Government Gates: https://helloai.com/articles/government-gated-frontier-releases (2026-07-01)
- Grok 4.3 Lands on Amazon Bedrock: https://helloai.com/articles/grok-4-3-now-on-amazon-bedrock (2026-06-22)
- Best Model for Coding Right Now: https://helloai.com/articles/best-model-for-coding-right-now (2026-06-22)
- Fable 5 Suspended: Government Directive Removes Mythos from All Users: https://helloai.com/articles/fable-5-suspended-us-government-directive (2026-06-13)
- Claude Fable 5: Mythos Power for General Use: https://helloai.com/articles/claude-fable-5-mythos-power-general-use (2026-06-10)
- Qwen3.7-Max Joins the Tracked Frontier: https://helloai.com/articles/qwen-3-7-max-joins-the-tracked-frontier (2026-06-04)
- The Entire Frontier Now Fits in Nineteen Elo Points: https://helloai.com/articles/the-frontier-is-a-near-tie (2026-05-30)
- Claude Opus 4.8: The Upgrade Is the Workflow, Not the Model: https://helloai.com/articles/claude-opus-4-8-the-orchestration-upgrade (2026-05-28)
- Most Agentic AI Projects Are Already Failing: https://helloai.com/articles/agentic-ai-is-failing-in-production (2026-05-26)
- Mythos: The First Frontier Model Too Dangerous to Ship: https://helloai.com/articles/mythos-too-dangerous-to-ship (2026-05-25)
- GLM-4.6: The MIT-Licensed Model Closing the Open Gap: https://helloai.com/articles/glm-46-the-mit-licensed-model-closing-the-open-gap (2026-05-22)
- Gemini 3.5 Flash: Faster, Cheaper, and Beating 3.1 Pro: https://helloai.com/articles/gemini-3-5-flash-faster-cheaper-and-beating-3-1-pro (2026-05-22)
- DeepSeek V4: The Open-Source Model Frontier Labs Feared: https://helloai.com/articles/deepseek-v4-open-source-frontier-parity (2026-05-15)
- Gemini 3.1 Pro Review: The Reasoning Leader You Haven't Tested: https://helloai.com/articles/gemini-3-1-pro-arc-agi-reasoning-review (2026-05-11)
- Google Just Bet $40B on Anthropic — What That Means for Your Stack: https://helloai.com/articles/google-40b-anthropic-bet (2026-05-05)
- Grok 4.3 Is Now the Cheapest Frontier Model: https://helloai.com/articles/grok-4-3-new-cost-leader (2026-05-03)
- GPT-5.5 "Spud" Doubles Its Price — And Bets Agents Are Worth It: https://helloai.com/articles/gpt-5-5-spud-doubles-its-price (2026-04-24)
- Claude Opus 4.7 and the End of the Frontier Cost War: https://helloai.com/articles/claude-opus-4-7-end-of-the-cost-war (2026-04-16)
- Claude Opus 4.6 Is Now 67% Cheaper — What Changes for Your Stack: https://helloai.com/articles/claude-opus-4-6-price-cut-what-it-means-for-your-stack (2026-04-15)
- The Advisor Strategy: How Anthropic Is Rethinking Model Costs: https://helloai.com/articles/the-advisor-strategy-how-anthropic-is-rethinking-model-costs (2026-04-10)
- Claude Mythos: The AI Too Dangerous to Release: https://helloai.com/articles/claude-mythos-the-ai-too-dangerous-to-release (2026-04-10)

## Related
- [HelloAI Marketplace](https://app.helloai.com): our own product, run by the same team as this site. Its MCP server is per-account; the public Hello, AI MCP server is listed above.