62 models from 9 providers, per million tokens. Cheapest right now: Llama 3.1 8B at $0.020 in / $0.040 out. Sort any column, or filter by provider and tier.
Lightweight 2.5 variant and the cheapest first-party rate tracked, but closed to new users. Native multimodal input. Cached input $0.01. Since Sep 18, 2026 Google limits the 2.5 models to developers who have used them before. Google deprecated 2.5 Pro and 2.5 Flash on Nov 2, 2026, but says 2.5 Flash-Lite is not deprecated and remains available through the API until further notice. New projects should start on 3.5 Flash-Lite or 3.8 Flash.
Budget flagship (Oct 7, 2026), replacing Haiku 4.5 at a tenth of its price. Rates shown are for prompts up to 100K tokens: a prompt over 100K bills the whole request at $0.50/$2.50, and cache reads and writes count toward the threshold. Cached read $0.01 (0.1x input), 5m cache write $0.125, 1h cache write $0.20, all 5x above 100K. Batch $0.05/$0.25 ($0.25/$1.25 above 100K). Ties GPT-6 Luna's $0.10/$0.50 on short prompts. Newer tokenizer counts the same text as ~30% more tokens than Haiku 4.5. 128K max output (300K on the Batch API, beta). Adaptive thinking, default effort medium.
Budget tier (Sep 22, 2026) and OpenAI's cheapest current-generation model, half the price of GPT-5.6 Luna on input. OpenAI's named replacement for GPT-5.4 Nano. Cached input $0.01, cache write $0.125. $0.20/$0.75 above 272K. 128K max output. Claude Haiku 5.5 (Oct 7) matches $0.10/$0.50, but only on prompts up to 100K tokens.
Meta's efficient execution model on the Meta Model API, built for high-volume sub-agent tasks, tool calls and routing. Listed here since Nov 10, 2026; Meta's pages give no launch date. Rates shown are for requests up to 200K input tokens: a request over 200K, cached tokens included, bills the whole request at $0.30/$1.50. Cached input $0.004 ($0.012 above 200K). Max output 524,288 tokens. Standard mode only: no Contributor tier and no batch discount.
Specialized for code generation (codestral-2508). Price cut in May 2026. 128K context. Mistral's only dedicated coding model since Devstral 2 retired on Jul 31, 2026.
Previous budget tier (GA Jul 9, 2026). Cut 80% from $1.00/$6.00 on Jul 30, 2026. GPT-6 Luna ($0.10/$0.50, Sep 22) now undercuts it on both input and output. Cached read $0.02, cache write $0.25 (1.25x input). 2x in/1.5x out above 272K.
Listed at the peak rate. Since Aug 16, 2026 DeepSeek bills peak/off-peak: off-peak is half ($0.15/$0.60), and peak runs 01:00-04:00 and 06:00-10:00 UTC on weekdays (09:00-12:00 and 14:00-18:00 Beijing time); all other hours are off-peak, including weekends and Chinese public holidays in full. V4.1 Flash (Sep 10, 2026) replaced V4 Flash as deepseek-flash; the old model names route to it. Cache hits $0.006 peak / $0.003 off-peak. Native vision.
Perplexity's own model, now billed through its Agent API (cached input $0.0625). Web search is a separate tool charge of $2.50 per 1,000 calls ($1 with Fast Search), replacing the old per-request fee. Sonar Chat Completions support ended Sep 27, 2026: synchronous and streaming calls are rewritten as Agent API requests, and asynchronous calls no longer work. Sonar Pro, Reasoning Pro and Deep Research map to Agent API effort levels (called presets until Oct 2026) that bill the underlying OpenAI or Anthropic model's rates.
Newest Mistral Large, a public preview since Oct 6, 2026, when it launched with a two-week 50%-off launch discount. General-purpose multimodal Mixture-of-Experts model with a 1M context window; Mistral says open weights are coming soon. Cached input $0.07.
Newest GA Flash (Sep 2, 2026), Google's most intelligent Flash model. Introductory $0.75/$3.75 through Dec 31, 2026; list price $1.50/$7.50 from Jan 1, 2027. Cached input $0.075 (list $0.15). Google warns it can use more tokens on long, complex tasks by design.
Listed at the peak rate after DeepSeek's Aug 16, 2026 price rise; off-peak is half ($0.66/$1.98) and covers weekends and Chinese public holidays in full. Cache hits $0.044 peak / $0.022 off-peak. A planned Sep 14 retirement was reversed, so V4 Pro stays available at unchanged billing.
Newest Terra model (Oct 26, 2026), which OpenAI pitches for visual reasoning, computer use and writing. Cached input $0.05 (5% of input), cache write $1.25. $2/$7.50 above 272K. 128K max output. Tool calling requires the Responses API.
Latest Muse Spark (Sep 2, 2026) on the Meta Model API, Meta's first paid first-party API. Same $1.25/$4.25 as 1.1 (Jul 9) and 1.2 (Aug 5), which stay available. Cached input $0.15. Multimodal agentic model with computer use. A Contributor tier bills $0.10/$0.20 (cached $0.002) in exchange for Meta training on your prompts and outputs, capped at 150 requests a minute. Still a public preview, but no longer US-only: access widened beyond the US with 1.2 on Aug 5.
Gemini 4 model for everyday coding and agentic workflows (GA Nov 12, 2026), which Google says brings cutting-edge intelligence to everyone. Cached input $0.075, implicit caching only. On the free tier. Adds an xhigh thinking level alongside low, medium and the default high.
Flagship (Sep 21, 2026), xAI's frontier model for coding, agentic tasks and knowledge work. Same $2/$6 and $0.50 cached input as Grok 4.6. $4/$12 above 200K prompt tokens. Batch 20% off. No output limit. Reasoning effort low to xhigh, default high. Also served on xAI's US regional endpoint at 1.1x the global rates.
Standard flagship (Sep 28, 2026), replacing Sonnet 5 at the same $2/$10. Cached read $0.10 (0.05x input, the same multiplier as Opus 5.5) since Oct 7, 2026, when Anthropic halved it from $0.20; cache writes unchanged at $2.50 (5m) and $4.00 (1h). 128K max output (300K on the Batch API, beta). Same tokenizer as Sonnet 5. Adaptive thinking, default effort high. No fast mode.
Mid tier (Sep 29, 2026), which OpenAI pitches as near-Astra performance at a fifth of the price. Replaced GPT-6 Sol (Sep 22, also $2/$10) a week after launch. Cached input $0.10, 5% of input where most OpenAI models charge 10%. Cache write $2.50. $4/$15 above 272K. 128K max output.
Google's most capable reasoning model (GA Nov 10, 2026), built for agentic coding, multimodal knowledge work, scientific reasoning and cybersecurity defense. Introductory $2/$10 through Nov 30, 2026; list price $4/$20 from Dec 1, 2026. Cached input $0.10 (list $0.20), implicit caching only. Not on the free tier.
Google's efficient Gemini 4 frontier reasoning model (GA Oct 13, 2026), which Google says gives faster responses and lower token spend per turn than Gemini 4 Carbon. Introductory $2/$10 through Nov 30, 2026; list price $4/$20 from Dec 1, 2026. Cached input $0.10 (list $0.20), implicit caching only. Not on the free tier.
Previous mid tier (GA Jul 9, 2026). Cut 20% from $2.50/$15.00 on Jul 30, 2026. GPT-6.1 Sol now matches its input price and undercuts it on output ($10 vs $12). Cached read $0.20, cache write $2.50 (1.25x input). 2x in/1.5x out above 272K.
New Opus flagship (Sep 22, 2026), 20% cheaper than Opus 5 on both input and output. Cache reads $0.20/MTok (0.05x input, half the standard Claude multiplier). Fast mode $8/$40. 128K max output; thinking is always on, default effort medium.
GA Jul 9, 2026. GPT-6.1 Sol ($2/$10) now sits between it and GPT-6 Astra at half its promotional price. Promotional $4/$20 since Aug 21, 2026 (list $5/$30), which OpenAI says runs until Nov 30, 2026. Cached read $0.40, cache write $5.00 (1.25x input). 2x in/1.5x out above 272K.
Current top tier (Nov 9, 2026), for demanding reasoning and long-horizon agentic work, at $8/$40 against Fable 5.1's $10/$50. Cache reads $0.20/MTok (0.025x input, the same multiplier as Fable 5.1). Batch $4/$20. Fast mode $16/$80. 128K max output; adaptive thinking is always on. Claude Mythos 5.2, which Anthropic says has the same capabilities, is limited to organizations verified through Anthropic's verification programs.
OpenAI's most capable model (Sep 3, 2026). Cached read $1.00, cache write $12.50 (1.25x input). $20/$75 above 272K. Fast mode bills 2x, and the Ultrafast tier (late Sep 2026) bills 6x ($60/$300). 128K max output. No 'none' reasoning effort or custom temperature/top_p, and tool calling requires the Responses API.
Foundry Global Standard (GA Sep 3, 2026), matching OpenAI direct. Cached input $1.00, cache write $12.50. $20/$75 above 272K. Data Zone deployments cost 10-20% more ($11/$55 to $12/$60, depending on zone). No batch pricing on Azure yet.
675B open-weight MoE, Mistral's flagship until Mistral Large 4 arrived in public preview on Oct 6, 2026. Context expanded from 32K to 256K. Deprecated Oct 31, 2026: API retires on Apr 30, 2027, with Mistral Large 4 as Mistral's named alternative.
Output price includes thinking tokens. Cached input $0.03. Since Sep 18, 2026 Google limits the 2.5 models to developers who have used them before. Deprecated Nov 2, 2026, with a Dec 1, 2026 shutdown date: from then Google routes gemini-2.5-flash requests to Gemini 3.5 Flash-Lite, billed at 3.5 Flash-Lite rates. New projects should start on 3.5 Flash-Lite.
Previous GA Flash (July 21, 2026), superseded by 3.7 and 3.8 Flash. Halved to an introductory $0.75/$3.75 in mid-August 2026, running through Dec 31, 2026; the $1.50/$7.50 list price returns Jan 1, 2027. Cached input $0.075 (list $0.15). Since Oct 8, 2026 Google routes requests for the retired Gemini 3.5 Flash to it.
Reasoning model on Azure. Deprecated: Azure retires it on Nov 19, 2026 (a month after OpenAI's own API shutdown on Oct 23), with GPT-5.6 Terra as the recommended replacement.
Previous budget model, now legacy after Haiku 5.5 replaced it on Oct 7, 2026. Haiku 5.5 is cheaper at every prompt length: a tenth of the price up to 100K tokens and half above. Cached read $0.10, 5m cache write $1.25, 1h cache write $2.00. 64K max output. On the older tokenizer, so the same text counts as fewer tokens than on Haiku 5.5.
Previous flagship (Aug 12, 2026), superseded by Grok 4.7 at the same $2/$6. Same token rates as Grok 4.5, but cached input is $0.50 (4.5 charges $0.30). $4/$12 above 200K prompt tokens. Batch 20% off. Also served on xAI's US regional endpoint at 1.1x the global rates.
Earlier flagship (Jul 8, 2026), superseded by Grok 4.6 and then 4.7, all at the same $2/$6. $4/$12 above 200K prompt tokens. Cached input $0.30. Batch 20% off.
Frontier-class multimodal model for agentic and coding work (v26.04). Open weights under a modified MIT licence. Deprecated Oct 31, 2026: API retires on Apr 30, 2027, with Mistral Large 4 as Mistral's named alternative.
Earlier GA flagship. 1M context. $2.50/$15 above 200K tokens. Cached input $0.125. Since Sep 18, 2026 Google limits the 2.5 models to developers who have used them before. Deprecated Nov 2, 2026, with a Dec 1, 2026 shutdown date: from then Google routes gemini-2.5-pro requests to Gemini 3.8 Flash, billed at 3.8 Flash rates. New projects should start on 3.8 Flash.
Previous Standard flagship, now legacy after Sonnet 5.5 replaced it on Sep 28, 2026 at the same $2/$10. That launch rate became the standard price on Aug 10, 2026, when Anthropic cancelled the scheduled Sep 1 rise to $3/$15. Cached read $0.20, 5m cache write $2.50, 1h cache write $4.00. New tokenizer bills ~30% more tokens than Sonnet 4.6 (1.0-1.35x). Full 1M context at standard rates.
Older Sonnet, superseded by Sonnet 5 and now Sonnet 5.5. Still strong for production workloads, and on the older tokenizer, so the same text counts as fewer tokens than on Sonnet 5.x.
Previous Opus flagship (July 24, 2026), superseded by Opus 5.5 at $4/$20. Launched at the same $5/$25 as Opus 4.8. Cached read $0.50. Fast mode $10/$50. Prompt-cache minimum 512 tokens.
Earlier Opus flagship (May 2026), superseded by Opus 5 at the same $5/$25 price and now legacy. Cached read $0.50. Fast mode $10/$50 until Nov 30, 2026, when Anthropic removes it (deprecated Oct 30, 2026).
Previous top tier (Sep 1, 2026), replacing Fable 5 at the same $10/$50, and now legacy since Fable 5.2 (Nov 9, 2026) arrived at $8/$40. Cache reads cut 75% to $0.25/MTok (0.025x input, as on Fable 5.2, against 0.05x on Opus 5.5 and Sonnet 5.5 and 0.1x on the other Claude models). 128K max output; thinking is always on.
Previous top tier (June 9, 2026), now legacy after Fable 5.1 replaced it at the same $10/$50. Cached read $1.00, four times Fable 5.1's rate.
No models match those filters
Blended = 3 parts input to 1 part output, a typical chat mix. Change = blended price, ▲ rise · ▼ cut since the October 2026 check; Mixed = input and output moved in opposite directions.Data CC BY 4.0
Price per million tokens, cheapest to dearest
Log scale by default, so the gaps between cheap models stay readable. Switch to linear, to the 15 cheapest, or to input or output price.
$0.01$0.10$1$10$100/MTok
Llama3.18B$0.025
Llama4Scout$0.15
Llama3.370B$0.16
Gemini2.5Flash-Lite$0.18
ClaudeHaiku5.5$0.20
GPT-6Luna$0.20
MuseEmber$0.20
MistralSmall4$0.26
Llama4Maverick$0.42
Codestral$0.45
GPT-5.6Luna$0.45
DeepSeekV4.1Flash$0.53
Sonar$0.81
Gemini3.5Flash-Lite$0.85
MistralLarge4$1.05
GrokBuild0.1$1.25
Gemini3.8Flash$1.50
Grok4.3$1.56
Grok4.20$1.56
GPT-5.4Mini$1.69
DeepSeekV4Pro$1.98
GPT-6.1Terra$2.00
MuseSpark1.3$2.00
Gemini4Hydrogen$3.00
Grok4.7$3.00
ClaudeSonnet5.5$4.00
GPT-6.1Sol$4.00
Gemini4Carbon$4.00
Gemini4Argon$4.00
GPT-5.6Terra$4.50
Gemini3.1ProPreview$4.50
GPT-5.4$5.63
ClaudeOpus5.5$8.00
GPT-5.6Sol$8.00
ClaudeOpus4.7$10.00
GPT-5.5$11.25
ClaudeFable5.2$16.00
GPT-6Astra$20.00
GPT-5.5Pro$67.50
Highlighted: the cheapest model in each tier. Current models only. Legacy models, and Azure OpenAI models (most share OpenAI’s list prices), are in the table but left out here. Price history over time →