ModelsPrice vs value
Price vs value
A current snapshot of frontier list prices next to what you get for each tier — context, modalities, reasoning notes, and the Artificial Analysis Intelligence Index already used on Models. Grouped by lab so Flash / Pro / Opus / Sol / Luna (and the rest) scan as a lineup, not a history chart.
For chat and product subscription tiers (Free / Plus / Pro / Enterprise), use Plan tiers. For price over time, use Pricing history. Human preference: Arena. Objective bench: LiveBench. Figures as of Sep 21, 2026. Missing cells are dashes, not guesses. Cached and batch rates appear only when publicly listed.
Disclaimer: List prices change. This is a sourced comparison for readers, not investment or purchasing advice. Prefer each lab's official pricing link over the summary cell.
Showing 28 of 28 tiers · snapshot Sep 1, 2026
Alibaba / Qwen
Official pricing →Qwen3.5 397B A17B Int4
3.5 Int4
—/—per 1M
in / out
- Context
- 262k
- AA Index
- 34
- Reasoning: reasoning (base model AA score)
- Text
- AA Index 34
Cheaper local serving of Qwen3.5 397B via GPTQ Int4 weights.
Self-host GPTQ Int4. Alibaba Cloud API for the base family is $0.60/$3.60 per 1M.
AA Index is base reasoning model, not a separate int4 eval. No first-party int4 token price.
Anthropic
Official pricing →Claude Fable 5.1
Fable
$10/$50per 1M
in / out
- Context
- 1M
- AA Index
- 66
- Reasoning: adaptive reasoning, max effort
- Text
- Vision
- Tools
- AA Index 66
- Cache $0.25/1M
Top-tier coding and research when you want the current AA leader.
Cache reads $0.25 per 1M (0.025× input). List unchanged from Fable 5.
Sep 2026 flagship; identical weights to Mythos 5.1.
Claude Mythos 5.1
Mythos
$10/$50per 1M
in / out
- Context
- 1M
- AA Index
- 66
- Reasoning: adaptive reasoning, max effort (scored via Fable 5.1 weights)
- Text
- Vision
- Tools
- AA Index 66
- Cache $0.25/1M
Same quality as Fable with more permissive cyber/life-sciences safeguards — if invited.
Invite-only — Project Glasswing / trusted access
Cache reads $0.25 per 1M where invited.
Not public API access. AA Index from shared Fable 5.1 weights.
DeepSeek
Official pricing →DeepSeek V4 Pro 0813
Pro
$1.32/$3.96per 1M
in / out
- Context
- 1M
- AA Index
- 53
- Reasoning: max effort
- Text
- Tools
- AA Index 53
Open-weight quality cut (1.6T MoE / 49B active) on MIT weights.
First-party API as measured by Artificial Analysis (may differ from off-peak list).
AA-measured first-party list.
DeepSeek V4 Flash 0731
Flash
$0.44/$1.32per 1M
in / out
- Context
- 1M
- AA Index
- 52
- Reasoning: max effort
- Text
- Tools
- AA Index 52
Cheaper DeepSeek V4 that still clears 50 on the Index.
First-party API as measured by Artificial Analysis.
AA-measured first-party list.
DeepSeek V4.1 Flash
Flash
$0.30/$1.20per 1M
in / out
- Context
- 1M
- AA Index
- 40
- Reasoning: max effort
- Text
- Image
- Tools
- AA Index 40
Fast open-weight Flash with image input; coding/agent-heavy workloads.
First-party API as measured by Artificial Analysis. AA Index 40 is v4.3 — not on the same scale as older v4.1.1 rows.
AA Intelligence Index v4.3 (max effort). Released 10 Sep 2026.
Google
Official pricing →Gemini 3.8 Flash
Flash
$0.75/$3.75per 1M
in / out
- Context
- 1.05M
- AA Index
- 41
- Reasoning: high
- Text
- Vision
- Audio
- Tools
- AA Index 41
- Cache $0.07/1M
Google’s Sep 2026 Flash flagship for agents and long-horizon coding at Flash prices.
Introductory through 31 Dec 2026; then $1.50 / $7.50 from 1 Jan 2027. AA Index 41 is v4.3.
AA Intelligence Index v4.3 (high).
Gemini 3.7 Flash
Flash
$0.75/$3.75per 1M
in / out
- Context
- 1M
- AA Index
- 56
- Reasoning: high
- Text
- Vision
- AA Index 56
High-throughput Gemini — currently outscores 3.1 Pro Preview on AA.
AA Google API listing; Google’s own Gemini 3.7 pricing page did not load when Models was last refreshed.
Confirm on Google pricing; Models page notes first-party page gaps.
Muse Spark 1.3
Muse Spark
$1.25/$4.25per 1M
in / out
- Context
- 1M
- AA Index
- 48
- Reasoning: max
- Text
- Vision
- Tools
- AA Index 48
- Cache $0.15/1M
Meta’s current API quality line (not Llama weights) for long-horizon coding and agents.
Standard muse-spark-1.3 route. AA Index 48 is v4.3 — not comparable to v4.1.1 rows.
AA Intelligence Index v4.3 (max). Proprietary Meta API.
Mistral
Official pricing →Moonshot / Kimi
Official pricing →OpenAI
Official pricing →GPT-6 Astra
Astra
$10/$50per 1M
in / out
- Context
- 1.05M
- AA Index
- 61
- Reasoning: max effort
- Text
- Vision
- Tools
- AA Index 61
- Cache $1/1M
OpenAI GPT-6 flagship for hard multi-step agent work.
Trusted Access first; Plus/Pro/Business/Enterprise rolling out.
Short-context list. Long context $20 / $75. Cache hit $1 short / $2 long.
AA Index 61 (max); xhigh 61; low 57.
GPT-5.6 Sol
Sol
$4/$20per 1M
in / out
- Context
- 1.05M
- AA Index
- 61
- Reasoning: max effort
- Text
- Vision
- Tools
- AA Index 61
OpenAI flagship quality; gpt-5.6 alias routes here.
Promotional list through at least 21 Nov 2026. Prompts over 272k input bill 2× input / 1.5× output.
GPT-5.6 Terra
Terra
$2/$12per 1M
in / out
- Context
- 1.05M
- AA Index
- 57
- Reasoning: max effort
- Text
- Vision
- Tools
- AA Index 57
Mid GPT-5.6 tier — most of Sol’s window at half the input price.
GPT-5.6 Luna
Luna
$0.20/$1.20per 1M
in / out
- Context
- 1.05M
- AA Index
- 52
- Reasoning: max effort
- Text
- Vision
- Tools
- AA Index 52
Volume GPT-5.6 — same 1.05M window, priced for throughput.
Grok 4.7
Flagship
$2/$6per 1M
in / out
- Context
- 500k
- AA Index
- 46
- Reasoning: xhigh
- Text
- Vision
- Tools
- AA Index 46
- Cache $0.50/1M
Current xAI coding and long-running agent flagship at the same list price as Grok 4.6.
Standard band under 200k prompt tokens ($2/$0.50 cached/$6). Over 200k: $4/$1/$12. Fast variant is 2× list. AA Index is v4.3.2 — not comparable to older v4.1.1 rows.
Released 21 Sep 2026. Available via API, Cursor, Grok Build. AA Intelligence Index 46 on v4.3.2 at xhigh.
Grok 4.6
Prior flagship
$2/$6per 1M
in / out
- Context
- 500k
- AA Index
- 61
- Reasoning: high effort
- Text
- Tools
- AA Index 61
Prior xAI coding/agent flagship on a 500k window; prefer Grok 4.7 for the current cut.
Standard band under 200k prompt tokens, per xAI docs.
Superseded as current flagship by Grok 4.7 (21 Sep 2026); same list $/1M on the standard band.
Zhipu / Z.AI
Official pricing →
Alibaba / Qwen
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Qwen3.8 2.4T A95B 3.8 Flagship | $2/$6per 1M in / out | 984k | Alibaba’s 2.4T MoE open-weight flagship (95B active).
| Source |
Qwen3.5 397B A17B Int4 3.5 Int4 | —/—per 1M in / out Self-host GPTQ Int4. Alibaba Cloud API for the base family is $0.60/$3.60 per 1M. | 262k | Cheaper local serving of Qwen3.5 397B via GPTQ Int4 weights.
AA Index is base reasoning model, not a separate int4 eval. No first-party int4 token price. | Source |
Anthropic
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Claude Fable 5.1 Fable | $10/$50per 1M in / out Cache reads $0.25 per 1M (0.025× input). List unchanged from Fable 5. | 1M | Top-tier coding and research when you want the current AA leader.
Sep 2026 flagship; identical weights to Mythos 5.1. | Source |
Claude Mythos 5.1 Mythos | $10/$50per 1M in / out Cache reads $0.25 per 1M where invited. | 1M | Same quality as Fable with more permissive cyber/life-sciences safeguards — if invited.
Invite-only — Project Glasswing / trusted access Not public API access. AA Index from shared Fable 5.1 weights. | Source |
Claude Fable 5 Fable (prior) | $10/$50per 1M in / out | 1M | Prior Fable cut; still listed for heaviest agent jobs at the top price band.
Superseded on quality by Fable 5.1; same list band. | Source |
Claude Opus 5 Opus | $5/$25per 1M in / out | 1M | Long-horizon agents and deep reasoning at half Fable’s list price.
| Source |
Claude Sonnet 5 Sonnet | $2/$10per 1M in / out | 1M | Default Claude workhorse — 1M context at mid-tier dollars.
| Source |
Claude Haiku 4.5 Haiku | $1/$5per 1M in / out | 200k | Fast, cheap Claude for lighter classification and short jobs.
Still on the current Anthropic price card; smaller window than the 5.x line. | Source |
DeepSeek
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
DeepSeek V4 Pro 0813 Pro | $1.32/$3.96per 1M in / out First-party API as measured by Artificial Analysis (may differ from off-peak list). | 1M | Open-weight quality cut (1.6T MoE / 49B active) on MIT weights.
AA-measured first-party list. | Source |
DeepSeek V4 Flash 0731 Flash | $0.44/$1.32per 1M in / out First-party API as measured by Artificial Analysis. | 1M | Cheaper DeepSeek V4 that still clears 50 on the Index.
AA-measured first-party list. | Source |
DeepSeek V4.1 Flash Flash | $0.30/$1.20per 1M in / out First-party API as measured by Artificial Analysis. AA Index 40 is v4.3 — not on the same scale as older v4.1.1 rows. | 1M | Fast open-weight Flash with image input; coding/agent-heavy workloads.
AA Intelligence Index v4.3 (max effort). Released 10 Sep 2026. | Source |
| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Gemini 3.8 Flash Flash | $0.75/$3.75per 1M in / out Introductory through 31 Dec 2026; then $1.50 / $7.50 from 1 Jan 2027. AA Index 41 is v4.3. | 1.05M | Google’s Sep 2026 Flash flagship for agents and long-horizon coding at Flash prices.
AA Intelligence Index v4.3 (high). | Source |
Gemini 3.7 Flash Flash | $0.75/$3.75per 1M in / out AA Google API listing; Google’s own Gemini 3.7 pricing page did not load when Models was last refreshed. | 1M | High-throughput Gemini — currently outscores 3.1 Pro Preview on AA.
Confirm on Google pricing; Models page notes first-party page gaps. | Source |
Gemini 3.1 Pro Preview Pro | —/—per 1M in / out No clean input/output pair independently confirmed on Google’s pricing page. | 1M | Pro-class preview; Flash 3.7 now leads it on this Index.
List $/1M unknown on this table. | Source |
Gemma 4 31B Gemma | —/—per 1M in / out Self-host / no first-party list price (AA lists $0/$0). | 256k | Apache-2 dense open weights when you self-host.
Not a Google API list-price tier. | Source |
| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Muse Spark 1.3 Muse Spark | $1.25/$4.25per 1M in / out Standard muse-spark-1.3 route. AA Index 48 is v4.3 — not comparable to v4.1.1 rows. | 1M | Meta’s current API quality line (not Llama weights) for long-horizon coding and agents.
AA Intelligence Index v4.3 (max). Proprietary Meta API. | Source |
Muse Spark 1.2 Muse Spark | $1.25/$4.25per 1M in / out | 1.05M | Historical Meta API quality cut on Index v4.1.1; superseded by Muse Spark 1.3.
Prior Meta API scorer on AA Index v4.1.1; prefer Muse Spark 1.3 for current quality. | Source |
Llama 4 Maverick Llama 4 | $0.26/$0.91per 1M in / out Median provider price on AA — not a first-party Meta API list. | 1M | Downloadable Llama-branded open weights; far behind Muse Spark on AA.
Hosted $/1M is third-party median. | Source |
Mistral
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Mistral Medium 3.5 Medium | $1.50/$7.50per 1M in / out | 256k | Mistral’s current mid-size open-weight sweet spot on this Index.
Large 3 scores well below Medium 3.5 on AA. | Source |
Moonshot / Kimi
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Kimi K3 K3 | $3/$15per 1M in / out | 1.05M | Highest-ranked open-weight on the current Index (2.8T MoE / 104B active).
Kimi K3 License (commercial with restrictions). | Source |
OpenAI
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
GPT-6 Astra Astra | $10/$50per 1M in / out Short-context list. Long context $20 / $75. Cache hit $1 short / $2 long. | 1.05M | OpenAI GPT-6 flagship for hard multi-step agent work.
Trusted Access first; Plus/Pro/Business/Enterprise rolling out. AA Index 61 (max); xhigh 61; low 57. | Source |
GPT-5.6 Sol Sol | $4/$20per 1M in / out Promotional list through at least 21 Nov 2026. Prompts over 272k input bill 2× input / 1.5× output. | 1.05M | OpenAI flagship quality; gpt-5.6 alias routes here.
| Source |
GPT-5.6 Terra Terra | $2/$12per 1M in / out | 1.05M | Mid GPT-5.6 tier — most of Sol’s window at half the input price.
| Source |
GPT-5.6 Luna Luna | $0.20/$1.20per 1M in / out | 1.05M | Volume GPT-5.6 — same 1.05M window, priced for throughput.
| Source |
| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
Grok 4.7 Flagship | $2/$6per 1M in / out Standard band under 200k prompt tokens ($2/$0.50 cached/$6). Over 200k: $4/$1/$12. Fast variant is 2× list. AA Index is v4.3.2 — not comparable to older v4.1.1 rows. | 500k | Current xAI coding and long-running agent flagship at the same list price as Grok 4.6.
Released 21 Sep 2026. Available via API, Cursor, Grok Build. AA Intelligence Index 46 on v4.3.2 at xhigh. | Source |
Grok 4.6 Prior flagship | $2/$6per 1M in / out Standard band under 200k prompt tokens, per xAI docs. | 500k | Prior xAI coding/agent flagship on a 500k window; prefer Grok 4.7 for the current cut.
Superseded as current flagship by Grok 4.7 (21 Sep 2026); same list $/1M on the standard band. | Source |
Zhipu / Z.AI
Official pricing →| Tier | $/1M in · out | Context | What you get | Source |
|---|---|---|---|---|
GLM-5.3 Pro | $1.40/$4.40per 1M in / out | 1M | Zhipu open-weight frontier; ties Kimi K3 at 60 on AA.
| Source |
GLM-5.3-Flash Flash | $0.15/$0.50per 1M in / out Native text+image input. ~10× cheaper list than GLM-5.3. | 1M | Multimodal Flash cut — ties Terra / Muse Spark quality at much lower $/1M.
MIT weights; 320B MoE / 18B active. | Source |