Best AI models: tier lists by job
Coding: Claude Opus 5.5; Writing: Claude Opus 5.5; Reasoning & analysis: Claude Opus 5.5; Long documents & context: Claude Sonnet 5.5; Cheap bulk tasks: GPT-6 Luna; Best value overall: GPT-6.1 Sol.
Editorial S/A/B tiers for 10 popular LLMs across 6 jobs, with links to every model's price.
Last updated: (AA index figures re-read on Index v5; LMArena ranks last checked ) · This page is reviewed weekly.
Leaderboard update, 29 Oct 2026: Index v5
Artificial Analysis moved to Intelligence Index v5 on 27 Oct 2026. On its leaderboard, Gemini 4 Carbon (high) is listed first at 59, one point above Claude Opus 5.5 at 58. The settings differ, it is one index, and Carbon is not publicly available yet.
| Model (setting) | v5 score | AA cost per task | Access |
|---|---|---|---|
| Gemini 4 Carbon (high) | 59 | $7.76 | Not public yet: going to selected cyber defenders first. Not on Google's API pricing page. |
| Claude Opus 5.5 (max with fallback) | 58 | $10.04 | Public |
| GPT-6 Astra (max) | 57 | $6.24 | Public |
| GPT-6.1 Sol (max) | 55 | $1.27 | Public |
| Gemini 4 Argon (high) | 50 | $7.68 | Public |
- Carbon access: Google says Carbon is going first to a subset of cyber defenders, then to more trusted partners through its Fairwind Program, with wider access later. It gives no date for that. Carbon is not on Google's Gemini API pricing page as of .
- Price: Google says Carbon will launch at $2 input / $10 output per 1M tokens, then move to $4 / $20 after an introductory period. Google gives no end date for that period. Artificial Analysis's cost figures use the standard $4 / $20 price, not the introductory one.
- Cost per task is Artificial Analysis's cost to run its own Index tasks at that setting. It is not a quote for your work. Model names and settings are copied from its leaderboard.
- Tier lists below: Carbon is not in them because it is not public. The "AA index" figures beside each model are Index v5 scores read on , and three tiers moved that day (see the changelog). LMArena ranks were last checked on ; we could not re-check them.
Top S-tier pick per job
Coding
Claude Opus 5.5
Writing
Claude Opus 5.5
Reasoning & analysis
Claude Opus 5.5
Long documents & context
Claude Sonnet 5.5
Cheap bulk tasks
GPT-6 Luna
Best value overall
GPT-6.1 Sol
Best for coding
Tier S
- Claude Opus 5.5
#1 in LMArena's agent leaderboard and the highest AA Index v5 score of the models we tier (58). The strongest pick for hard agentic coding.
- GPT-6.1 Sol
#2 in LMArena's agent leaderboard at a fraction of Opus's cost per task. Excellent for agent loops.
- Claude Sonnet 5.5
Within four points of Opus on AA Index v5 (54 vs 58) at half the price, with a full 1M context.
Tier A
- Gemini 4 Argon
#4 in LMArena's agent leaderboard. Google pitches it for complex software engineering.
- Grok 4.7
xAI's recommended model for code. Solid, but scores below the leaders on the AA index.
Tier B
- Claude Haiku 5.5
Fast and cheap for small edits and autocomplete-style tasks. Not for big refactors.
- Gemini 3.8 Flash
Google pitches it for long-horizon coding. A good budget agent, below the frontier models.
- DeepSeek Flash (V4.1)
Very cheap coding loops. A lower index than the frontier models.
- GPT-6 Luna
On LMArena's agent price/performance frontier. Best for simple, high-volume code tasks.
- DeepSeek V4 Pro
Scores below DeepSeek Flash on the AA index, so there's little reason to pick it for coding.
Best for writing
Tier S
- Claude Opus 5.5
Top overall intelligence. Our pick for long-form, nuanced writing and editing.
- Claude Sonnet 5.5
Near-Opus quality at half the price. Our default for everyday writing.
Tier A
- GPT-6.1 Sol
A strong all-rounder with low cost per task.
- Gemini 4 Argon
A strong frontier model. Good for research-heavy writing.
- Grok 4.7
Capable general writer, priced below the other flagships.
Tier B
- Claude Haiku 5.5
Good for short copy, summaries and rewrites at a low price.
- Gemini 3.8 Flash
Quick drafts and summaries.
- GPT-6 Luna
Cheap short-form copy.
- DeepSeek Flash (V4.1)
Cheap drafts. Edit before publishing.
- DeepSeek V4 Pro
Usable, but no clear edge over Flash.
Best for reasoning & analysis
Tier S
- Claude Opus 5.5
Highest AA Index v5 score of the models we tier (58). Only Gemini 4 Carbon, which is not public yet, is listed higher (59).
- GPT-6.1 Sol
55 on AA Index v5, second of the models we tier and above Sonnet (54), at $1.27 per AA task, the lowest of the flagships.
- Claude Sonnet 5.5
54 on AA Index v5 (max with fallback), third of the models we tier and within four points of Opus (58).
Tier A
- Gemini 4 Argon
Google's flagship reasoning model. 50 on AA Index v5 (high), behind Opus, Sol and Sonnet (54 to 58).
- Claude Haiku 5.5
A strong small model: 36 on AA Index v5 (max), above Grok 4.7 (34) and DeepSeek Flash (32).
- Grok 4.7
34 on AA Index v5 (xhigh): the lowest of the flagships, and close to the small models.
- DeepSeek Flash (V4.1)
Budget reasoning with thinking mode. 32 on AA Index v5 (max), close to Grok 4.7 (34).
Tier B
- Gemini 3.8 Flash
Decent reasoning for a Flash-tier model.
- GPT-6 Luna
Light analysis at minimal cost.
- DeepSeek V4 Pro
Scores below Flash on the AA index despite the higher price.
Best for long documents & context
Tier S
- Claude Sonnet 5.5
Full 1M context at the standard price (no long-context surcharge), with top-tier intelligence.
- Claude Opus 5.5
1M context at the standard price. Best quality for deep document analysis, but expensive at volume.
Tier A
- Gemini 4 Argon
1M context per Artificial Analysis. Strong reasoning across long inputs.
- GPT-6.1 Sol
1M context per Artificial Analysis, but prices rise above 272K input tokens.
- DeepSeek Flash (V4.1)
1M context confirmed on DeepSeek's own pricing page, at a very low price. Ideal for cheap long-document RAG.
Tier B
- Grok 4.7
500K context, and prices double at 200K+ prompt tokens.
- Claude Haiku 5.5
Prices are 5× higher for prompts over 100K tokens.
- Gemini 3.8 Flash
1M context per Artificial Analysis. A budget option, but less accurate.
- GPT-6 Luna
Prices rise above 272K input tokens. Fine for mid-size documents.
- DeepSeek V4 Pro
1M context, but no quality edge over Flash.
Best for cheap bulk tasks
Tier S
- GPT-6 Luna
$0.20 blended per 1M tokens, plus a 50% batch discount. On LMArena's agent price/performance frontier.
- Claude Haiku 5.5
$0.20 blended for prompts up to 100K tokens, plus batch. Scores 36 on AA Index v5 (max).
- DeepSeek Flash (V4.1)
$0.525 blended at peak and half that off-peak, with 1M context.
Tier A
- Gemini 3.8 Flash
Good quality per dollar at its promo price (it doubles on 1 Jan 2027).
- DeepSeek V4 Pro
Cheap for a large model, especially off-peak.
Tier B
- Grok 4.7
Too expensive for bulk work.
- GPT-6.1 Sol
Too expensive for bulk work unless you need its quality. Use batch.
- Claude Sonnet 5.5
Too expensive for bulk work. Use batch if you must.
- Gemini 4 Argon
Too expensive for bulk work, and the price doubles on 1 Dec 2026.
- Claude Opus 5.5
The most expensive model here.
Best for best value overall
Tier S
- GPT-6.1 Sol
Near-frontier intelligence (AA Index v5: 55) at $1.27 per AA task, far below the other flagships. On LMArena's agent price/performance frontier.
- Claude Haiku 5.5
An AA Index v5 score of 36 for $0.20 blended. The cheapest strong small model here.
Tier A
- Claude Sonnet 5.5
54 on AA Index v5, within four points of Opus, for a $4 blend. Premium quality at a mid price.
- DeepSeek Flash (V4.1)
An AA Index v5 score of 32 at a $0.525 peak blend (half off-peak), with 1M context. Haiku scores 36 for less.
- GPT-6 Luna
The cheapest model here. Lower quality, but unbeatable for simple tasks.
- Gemini 4 Argon
Strong value until its promo ends on 30 Nov 2026.
- Gemini 3.8 Flash
Good value until the promo ends on 31 Dec 2026.
Tier B
- Grok 4.7
Its AA Index v5 score of 34 trails cheaper or similar-priced options.
- Claude Opus 5.5
The best model we tier (v5: 58). Costs 2× Sonnet per token for a small gain, but AA's cost per task is about the same ($10.04 vs $10.02).
- DeepSeek V4 Pro
Scores lower than the cheaper DeepSeek Flash.
Priced but not tiered
These models are in our price table but in none of the tier lists above, because we have not given them a tier.
- Claude Fable 5.1 (Anthropic): $10.00 input / $50.00 output per 1M tokens ($20.00 blended), the highest price in the table. Anthropic lists it on the Claude API. It says to start with Claude Opus 5.5 for most workloads and to use Fable 5.1 for demanding reasoning and long-horizon agentic work.
- Claude Mythos 5.1 (Anthropic) Limited access: the same price as Fable 5.1, but Anthropic offers it only to organizations it has verified, such as through its Cyber Verification Program, so most readers cannot use it. It is not a recommendation.
How we rank models
Tiers are editorial opinion, not a single benchmark. For each job we weigh:
- Public leaderboards: the Artificial Analysis Intelligence Index (at the effort level shown) and LMArena's agent rankings.
- Official docs: context windows, long-context surcharges and what each vendor says the model is built for.
- Price: official per-token prices from our price comparison. Price weighs most for the bulk and value lists.
S = best for the job right now, A = excellent, B = usable with trade-offs. We never accept payment for tiers. Scores change often, so check the linked sources. The writing tiers draw on overall intelligence scores and our judgment, because there's no widely accepted writing benchmark.
Sources
- Artificial Analysis LLM Leaderboard (Intelligence Index v5, cost per task, context window), checked
- LMArena Leaderboard (Top 10 Agents, Pareto frontier), checked
- Official pricing and model docs (see /cheapest-ai-models/ for each URL), checked
- Artificial Analysis leaderboard, Index v5 (update section), checked
- Artificial Analysis: Gemini 4 Carbon is the new #1 on Index v5, checked
- Google: Introducing Gemini 4 Carbon, checked
- Google Gemini API pricing (no Carbon listed), checked
Changelog
- No tier changes. Added a short note that Claude Fable 5.1 and Claude Mythos 5.1 ($10 / $50 per 1M tokens each, from Anthropic's pricing page) are in the price table but have no tier. Mythos 5.1 is labelled Limited access. Tiers were last reviewed on 2 Nov.
- Weekly review: no tier changes. Artificial Analysis Index v5 figures re-checked and unchanged for all 10 tiered models; Gemini 4 Carbon is still not on Google's API pricing page, so it stays untiered. LMArena not re-checked (arena.ai returned 403); last checked 16 Oct. Llama 3.3 70B removed from all six lists because Together AI no longer offers it on its serverless API, so its price can't be verified.
- Re-scored on AA Index v5: GPT-6.1 Sol moved A to S for reasoning (55, above Sonnet's 54); Claude Haiku 5.5 moved B to A for reasoning (36, above Grok 4.7's 34 and DeepSeek Flash's 32); DeepSeek Flash moved S to A for best value (32, against Haiku's 36 at a lower price). All AA index figures are now v5 and 16 reasons were reworded. Grok 4.7 unchanged. LMArena ranks not re-checked (arena.ai returned 403); last checked 16 Oct.
- Added a dated Index v5 leaderboard section near the top. Gemini 4 Carbon is untiered because it is not public yet.
- DeepSeek Flash's AA Intelligence Index rose from 39 to 45 (now DeepSeek V4.1 Flash 1017), so it moved from B to A for reasoning; Claude Haiku 5.5 reasons no longer call it the top small model. Arena Top 10 Agents not rechecked (page returned 403).
- Weekly review: no tier changes
- Page launched: 6 job tier lists for 11 models.
FAQ
What is the best AI model for coding?
Claude Opus 5.5 and GPT-6.1 Sol and Claude Sonnet 5.5 are S-tier for coding in our 2 Nov 2026 review. #1 in LMArena's agent leaderboard and the highest AA Index v5 score of the models we tier (58). The strongest pick for hard agentic coding.
What is the best AI model for writing?
Claude Opus 5.5 and Claude Sonnet 5.5 are S-tier for writing in our 2 Nov 2026 review. Top overall intelligence. Our pick for long-form, nuanced writing and editing.
What is the best AI model for reasoning and analysis?
Claude Opus 5.5 and GPT-6.1 Sol and Claude Sonnet 5.5 are S-tier for reasoning & analysis in our 2 Nov 2026 review. Highest AA Index v5 score of the models we tier (58). Only Gemini 4 Carbon, which is not public yet, is listed higher (59).
What is the best AI model for long documents?
Claude Sonnet 5.5 and Claude Opus 5.5 are S-tier for long documents & context in our 2 Nov 2026 review. Full 1M context at the standard price (no long-context surcharge), with top-tier intelligence.
What is the best AI model for cheap bulk tasks?
GPT-6 Luna and Claude Haiku 5.5 and DeepSeek Flash (V4.1) are S-tier for cheap bulk tasks in our 2 Nov 2026 review. $0.20 blended per 1M tokens, plus a 50% batch discount. On LMArena's agent price/performance frontier.
What is the best value AI model overall?
GPT-6.1 Sol and Claude Haiku 5.5 are S-tier for best value overall in our 2 Nov 2026 review. Near-frontier intelligence (AA Index v5: 55) at $1.27 per AA task, far below the other flagships. On LMArena's agent price/performance frontier.
How are these tiers decided?
They are editorial. We combine official model docs with public leaderboards (Artificial Analysis, LMArena) and official prices. We review them weekly.