Fireworks Research unveiled Ember-1 on September 23 as a research preview, building upon Moonshot's open-weight Kimi K3 model. The company claims Ember-1 achieves comparable quality to Kimi K3 while consuming approximately 40% fewer tokens. According to Fireworks, the model "learned to cut unnecessary reasoning while keeping the thinking that matters." Since reasoning tokens are billed as output, models requiring less reasoning should theoretically reduce costs.

Pricing presents a complication. On OpenRouter, Ember costs $3 per million input tokens and $15 per million output tokens—identical to Fireworks' pricing for Kimi K3. However, other providers offer Kimi K3 at rates as low as $1 per million input tokens and $9 per million output tokens. Testing routed both models through OpenRouter on Fireworks to maintain pricing parity, while also calculating what Kimi would cost at lower-priced alternatives.

Accuracy matters more than cost savings if a model cannot deliver reliable results. Testing examined whether Ember's reduced reasoning maintained correctness, with the hypothesis that cutting necessary reasoning would first degrade performance on the most challenging problems. Tests progressed in difficulty and ran five times each to verify consistency.

Test methodology

Both models were called through OpenRouter with identical prompts and default reasoning settings, routed to Fireworks to ensure fair speed comparison. Each test executed five times per model, with reasoning tokens logged separately from other output.

  • Logic puzzles: Three puzzles of increasing complexity involving 4, 5, and 7 engineers with one solution each. Larger puzzles require longer reasoning chains, making them sensitive to reasoning cuts.
  • Deploy scheduling: Twelve services with dependencies, one deploy per team at a time, and two blackout windows. The model must identify the fastest possible rollout: 17 hours.
  • Probability: Five questions about a retry system with a server alternating between healthy and degraded states, plus a circuit breaker. All answers are exact fractions, verified against a simulation of 2 million requests.

Answer keys were confirmed using two independent methods before either model attempted the tests. Prompts appear at the end of this article for replication.

Results by category

Logic puzzles

Both models solved all three puzzles across all five runs. Ember averaged 13,630 reasoning tokens, 3 minutes 46 seconds, and $0.27 per run. Kimi averaged 16,679 reasoning tokens, 12 minutes 26 seconds, and $0.34. Ember used 18% fewer reasoning tokens and demonstrated significantly faster execution, with Kimi's slowest run approaching 20 minutes.

Deploy scheduling

Both models found the 17-hour schedule on every attempt, with all schedules passing validation. Ember averaged 6,543 reasoning tokens, 1 minute 29 seconds, and $0.10. Kimi averaged 7,792 reasoning tokens, 4 minutes 46 seconds, and $0.13. Ember achieved 16% fewer reasoning tokens and substantially faster performance.

Probability

This category produced the only error. Kimi answered all five questions correctly on every run. Ember achieved four perfect runs but made an arithmetic error on the fifth, calculating 0.94619 instead of 0.94629 on the first question, with the error propagating to two additional answers. Ember averaged 6,242 reasoning tokens, 1 minute 47 seconds, and $0.13. Kimi averaged 9,682 reasoning tokens, 6 minutes 48 seconds, and $0.19. This test came closest to Ember's 40% token reduction claim, and Ember again demonstrated significantly faster execution.

Overall performance

Kimi K3 achieved 15 perfect runs across all tests, while Ember-1 achieved 14. Ember's single miss—an arithmetic slip on the probability test—appears minor given it occurred in just one of 15 runs. A notable finding absent from marketing claims: Ember-1 completed each test set 3.4 times faster than Kimi K3.

Regarding reasoning token efficiency, Ember-1 used 23% fewer reasoning tokens than Kimi K3. On Fireworks' pricing, Ember-1 cost $2.48 total compared to Kimi K3's $3.26, representing a 24% savings. However, this advantage evaporates when Kimi K3 is accessed through cheaper providers. At the lowest available pricing of $1 per million input tokens and $9 per million output tokens, the same test runs would have cost $1.96, undercutting Ember-1.

Practical considerations

Ember-1 matches Kimi K3's accuracy while consuming substantially fewer reasoning tokens. Cost advantages on Ember-1 prove less decisive because users seeking Kimi K3 can route through cheaper OpenRouter providers, bringing pricing below Ember-1's levels. Kimi K3 is known for slow performance paired with low cost. Speed measurements derive from Fireworks' standard endpoint; cheaper providers were not tested. For latency-sensitive applications requiring near-identical results to Kimi K3, Ember-1 offers compelling advantages. For budget-conscious users with flexible timelines, Kimi K3 remains competitive when sourced from lower-cost providers.

Test prompts

Each prompt concludes with a fixed answer format enabling automatic grading.

Logic puzzles

Solve all three logic puzzles below. Each has exactly one solution.

PUZZLE SMALL: 4 engineers (Ava, Bo, Cleo, Dev) each own exactly one server. Each server has a rack position (1, 2, 3, 4, numbered left to right), an operating system (Debian, Alpine, Fedora, Ubuntu), a role (cache, queue, proxy, db). No two servers share any of these values.

  • If the server in rack 3 runs Fedora, then the Debian server is in rack 3.
  • The server in rack 1 is Dev's.
  • Exactly one of these is true: the server in rack 4 runs Ubuntu, or the Alpine server is not Bo's.
  • Cleo is the proxy.
  • The db server is Dev's.
  • Ava runs Alpine.
  • The queue server is not Ava's.
  • Bo is in rack 3.
  • Bo and the Ubuntu server are in neighboring racks.

PUZZLE MEDIUM: 5 engineers (Ava, Bo, Cleo, Dev, Eun) each own exactly one server. Each server has a rack position (1, 2, 3, 4, 5, numbered left to right), an operating system (Debian, Alpine, Fedora, Ubuntu, Arch), a role (cache, queue, proxy, db, build), a replication data center (Oslo, Lima, Pune, Accra, Perth). No two servers share any of these values.

  • Cleo is the db.
  • Exactly one of these is true: the cache server runs Arch, or the server replicated to Perth is the proxy.
  • The server in rack 5 runs Fedora.
  • Exactly one of these is true: the server in rack 2 is not the proxy, or Cleo runs Debian.
  • Exactly one of these is true: the server replicated to Pune is the db, or Dev is in rack 3.
  • The server in rack 1 is the db.
  • The proxy server is in rack 5.
  • The server in rack 4 runs Ubuntu.
  • The cache server and the server replicated to Pune are in neighboring racks.
  • Ava is the cache.
  • The server replicated to Accra is the queue.
  • The server replicated to Lima is not in rack 3.
  • Eun runs Alpine.
  • The queue server is not in rack 2.
  • The Fedora server is not Dev's.
  • The server in rack 2 is not replicated to Lima.

PUZZLE LARGE: 7 engineers (Ava, Bo, Cleo, Dev, Eun, Finn, Gia) each own exactly one server. Each server has a rack position (1, 2, 3, 4, 5, 6, 7, numbered left to right), an operating system (Debian, Alpine, Fedora, Ubuntu, Arch, Rocky, NixOS), a role (cache, queue, proxy, db, build, metrics, auth), a replication data center (Oslo, Lima, Pune, Accra, Perth, Quito, Riga). No two servers share any of these values.

  • The build server is Cleo's.
  • The proxy server does not run Fedora.
  • The server replicated to Quito is Finn's.
  • The server in rack 6 is replicated to Lima.
  • The Alpine server is exactly 4 racks to the right of the auth server.
  • The server replicated to Oslo is Dev's.
  • The Debian server is replicated to Quito.
  • The server replicated to Accra is not the build.
  • The Rocky server is exactly 1 rack to the right of Gia.
  • Exactly one of these is true: the Ubuntu server and the server replicated to Pune are in neighboring racks, or the server replicated to Oslo runs Fedora.
  • The server replicated to Oslo does not run Fedora.
  • Ava runs Rocky.
  • Exactly one of these is true: the proxy server is somewhere to the left of the server replicated to Riga, or the Debian server is in rack 7.
  • The auth server and Bo are in neighboring racks.
  • The metrics server is Gia's.
  • The Fedora server is somewhere to the left of the server replicated to Pune.
  • Exactly one of these is true: the server replicated to Accra is in rack 2, or Finn is the queue.
  • If the Debian server is Finn's, then the server in rack 7 does not run Rocky.
  • The NixOS server is not replicated to Perth.
  • The server in rack 7 is replicated to Accra.
  • The server replicated to Pune is the db.

At the end of your response, give each solution as a block, one line per engineer, in the order the engineers are listed in that puzzle.

Deploy scheduling

You are planning a production rollout of 12 services. Time is measured in whole hours from hour 0.

Services (owning team, deploy duration in hours): The prompt includes 12 services with specified owning teams and durations.

Dependencies: A service may start deploying only after every service it depends on has finished deploying.

Team constraints: Each team can deploy only one of its services at a time. Different teams can deploy simultaneously.

Blackout windows: No deploy may be in progress during hours 9 to 11 or hours 17 to 19. A deploy may end exactly at hour 9 or 17, and may start exactly at hour 11 or 19. A deploy cannot pause and resume. Each deploy runs from its start hour for its full duration without interruption.

What is the earliest hour by which all 12 services can be finished? Give a schedule that achieves it.

Probability and circuit breaker

A client calls a payment API and retries on failure. The server is in one of two states on each attempt, Healthy or Degraded.

On the first attempt, the server is Healthy with probability 4/5 and Degraded with probability 1/5. Between consecutive attempts, the state changes as follows: from Healthy, it stays Healthy with probability 3/4 and becomes Degraded with probability 1/4. From Degraded, it stays Degraded with probability 2/3 and becomes Healthy with probability 1/3.

An attempt succeeds with probability 9/10 if the server is Healthy and 2/5 if it is Degraded, independently of everything else given the state. The client makes at most 4 attempts and stops as soon as one succeeds.

Circuit breaker: If two consecutive attempts both hit a Degraded server and both fail, the client stops immediately and makes no more attempts.

Answer these questions with exact fractions in lowest terms:

  1. What is the probability that the request eventually succeeds?
  2. What is the expected number of attempts the client makes?
  3. Given that the request succeeds, what is the probability that it succeeded on exactly the second attempt?
  4. What is the probability that the circuit breaker ends the request early, before the client has used all 4 attempts?
  5. Given that the request succeeds, what is the probability that the first attempt failed?

At the end of your response, give exactly five lines in this format: Q1. [answer], Q2. [answer], Q3. [answer], Q4. [answer], Q5. [answer]